As much as I like Tildes and understand the premise that we should all be trying to consider each others' opinions, I'm saddened by how people on here seem to not only accept genAI programming so...
Exemplary
As much as I like Tildes and understand the premise that we should all be trying to consider each others' opinions, I'm saddened by how people on here seem to not only accept genAI programming so readily, but also completely dismiss ethics when justifying why they do so.
There's a few people on here implying this would only be relevant if it only excluded vibe-coded projects, not AI assistance. That only the resulting product matters. Did we forget everything we were criticizing large models for in the first place? Do we just not care that the tool itself is based on stolen work? That continued use and support of it is actively destroying the environment for little benefit? That its growing commodity is causing a general decrease in individual knowledge and learning capacity? That it's making the industry of IT that most of the people commenting belong to worse to be a part of? Like, surely, a bit more than just the result matters here?
Ethics matter. Call it a purity contest or virtue signaling or whatever, I don't care. Trying to do something good to push against the bullshit even if it's imperfect and still has to rely on tech that does use it to exist and be effective is still better than doing nothing at all. This is a good thing and I have no criticism to make of it.
Yo š I'm personally trying to tone down how much confrontation I cause in this space, but I wanted to minimally note my disapproval. Framing this as a black and white ethical issue equally saddens...
Exemplary
Yo š I'm personally trying to tone down how much confrontation I cause in this space, but I wanted to minimally note my disapproval. Framing this as a black and white ethical issue equally saddens me, since that same absence of nuance and appeal to morality has been used to justify atrocious actions throughout history. And it erases any good that the tools can accomplish. Furthermore, given the amount of personal attacks I get whenever I mention LLMs/AI here, I sincerely doubt that there's a majority of people who agree with me -- you are almost certainly in the majority. I'd imagine people who hate this have artificial intelligence filtered out of their tags, or otherwise avoid the topics.
(edit: sigh ... I had hoped this wouldn't be exemplary'd. For anyone reading this, fwiw, the person was pleased with the notion of trying to keep conversations calmer. It wasn't an AI thing)
I was going to give an exemplary label because of this, but I wanted to show publicly why I 100% agree with your point. I'm so, so, so tired of this 0 or 1 attitude. You're either 100% in favour...
Framing this as a black and white ethical issue equally saddens me
I was going to give an exemplary label because of this, but I wanted to show publicly why I 100% agree with your point.
I'm so, so, so tired of this 0 or 1 attitude. You're either 100% in favour of something, or 100% against it. It feels like nowadays on the internet you can't stand somewhere between both. This attitude I feel is more destructive in the long run.
I, for example, think:
Image and Video Gen AIs overall are more bad than good, because of the impact they have on real artists and the dangerous potential for misinformation (which I feel it's only going to get worse as these models get better)
Still, from a purely technical stand point, I'm blown away by these things. Think about it: we "taught" sand to make videos! How cool is that!?
These models were and are being trained without consent, and people are rightly pissed off about it. I'm not an artist but I want to make video games, I would also be pissed off as well if someone used my games without consent
But these models are a genie that are out of the bottle now. Are we supposed to pretend they don't exist? Should the americans 70 years ago pretended jet engines weren't a thing because the Nazis invented them? I don't think so.
I'm fatigued as well about almost every company wanting to shove AI into every single piece of software. It's nuts, it's stupid, I'm against it. It's a privacy nightmare as well and the wet dream for Google and others.
LLM's allowed me to finally make my own homelab and troubleshoot some niche things on Linux. I'm building something I always wanted to have at home, and I'm learning new skills as I go along.
On the other hand, I do feel the "temptation" of offloading thinking process to the model. I've been disciplined to always thinking of the solutions myself and let the LLM execute, but I imagine many people are not doing the same (be it in coding context or others). I worry if/when these companies decide to influence the public opinion, if they aren't already.
I have a lot more thoughts, but these zigzag of thoughts should be good enough to illustrate my point. No one will ever convince me that these things are the best thing ever - we shouldn't underestimate some of the societal dangers these could potentially have. But neither will no one ever convince me that my own experiences are wrong, and that LLMs didn't have some good outcomes for my own skills and work.
I haven't posted on here a lot about LLM stuff, but I keep reading most of it...or at least, most of the threads on it, which have not been great for me mentally. I think I'm fairly anti-ai, but I...
I haven't posted on here a lot about LLM stuff, but I keep reading most of it...or at least, most of the threads on it, which have not been great for me mentally.
I think I'm fairly anti-ai, but I have a hard time really examining my position because I know how emotionally biased I am in it.
I am not an artist. I have never been very "creative", but I love the thought of being creative.
I loved legos growing up, and I think most of that is because it let me put things together
in ways that were creative and showing creativity, but did not require me to be "good" at it.
I was never able to get into drawing or anything like that in the same way, and I've always
loved reading comics and seeing what artistic people are able to create.
I went to school for programming, because I truly love writing code & seeing things happen as a result.
To me, programming has been my "creative" output. So I find myself biased against LLMs generating code, for what I imagine are the same emotional reasons as a lot of artists, but not necessarily the reasons that get sympathy or could actually have legal repercussions (e.g. training on stolen art).
But also I struggle with it myself, becuase like, am I gatekeeping? Am I against it because it was my "special thing" that many others couldn't do? I don't think so, but maybe there's some of that to it as well, as to why I don't like it. But like, at the same time, I'm glad to hear that it has helped you be able to set up a homelab and troubleshoot stuff on Linux.
I think it is great that people are now able to customize software and make tools and stuff that are useful to them, that they would never have been able to "before LLMs". Except, if I look at this from the art angle, I'm not sure how consistent my view is here?
I've never really been good at putting my feelings into words, and I certainly haven't gotten any better at it. It just seems like most of what I have been good at, are things that now people are using LLMs for.
and I don't know where that leaves me.
I'm going to keep avoiding LLMs for the things I enjoy doing, which is...most of the things I do,
and it does frustrate me seeing LLMs continually invade those things, and it is sometimes saddening
when I see others using LLMs for the things I enjoy doing. I don't think others should necessarily
not use them for those things, because if they don't enjoy doing them, sure, automate it away.
But like, it is much, much too late to put the genie back into the bottle, so all I can really do is personally protest by avoiding using them where I can. And I don't know whether I'm doing that for ethical or emotional reasons, but I'm also not really sure it matters.
I do really appreciate your take here, as well as from @kacey. I like being able to have some of these nuanced conversations around LLM uses, (or at least, read other peoples nuanced takes when I don't feel like commenting).
Yeah, I think I'm pretty aligned with you on my feelings around this stuff. I don't mind when other people use AI to help them do personal projects, or when people use it at work to help with...
Yeah, I think I'm pretty aligned with you on my feelings around this stuff. I don't mind when other people use AI to help them do personal projects, or when people use it at work to help with work, as long as the quality is still good (big qualifier, I know).
I do absolutely HATE when people diminish the work of other people by commenting that AI could do that, or that it wasn't actually that hard because they did the same thing with AI. It fills me with blinding rage, and it's very common amongst obnoxiously pro ai spaces.
When someone hand builds a complicated piece of software, shares it with the world, talks about the years they've spent on it, and someone comments that AI could do the same thing, I just want to slap them.
Even if it was the same quality, which is extremely dubious, imagine someone going up to a master blacksmith that created a hand forged masterpiece of a sword over months and going "I could get the same thing on temu for 10 dollars".
This unfortunately happens (very common in custom furniture especially), but I absolutely detest the idea.
Even if the results were the same (they basically never are). Someone hand crafted by a human being is special, and will always be more valuable than something mass produced and shitted out by a machine. Even if blacksmithing and furniture making is no longer a viable career in most places, the people those things create has way more intrinsic value to most people, and the same is true for software in my opinion.
Thank you very much for not doing that š I feel that it needlessly escalates tensions on a topic that needs to be discussed (as noted in your comment!). For my part, I've been wishing for years...
I was going to give an exemplary label because of this [...]
Thank you very much for not doing that š I feel that it needlessly escalates tensions on a topic that needs to be discussed (as noted in your comment!).
For my part, I've been wishing for years that the temperature would come down far enough to talk about effective resistance against all this -- job loss via automation, devaluing of the arts, legislative reform, etc. But that requires acknowledging that the problem exists, and dispassionately analyzing the space for asymmetries to be exploited in favour of the common people.
We need to be organizing workers, pressuring local governments, and pushing our vision of what a future with these tools should look like -- not de-facto rolling over and allowing OpenAI/Anthropic/Microsoft/Google to decide for us. Endlessly debating about data centres and morality cannot un-create the idea of a transformer, or delete every open weight model on the internet, but I firmly believe that the window of opportunity remains open for deciding whether the world on the other side of all this is a good one. For whatever definition we -- not Sam Altman or Dario Amodei -- decide "good" is.
I see these questions often and they feel a little.. manipulative (perhaps unintentionally?). Because I've never seen anyone ask that we pretend LLMs don't exist. It's a straw man. Why do so many...
But these models are a genie that are out of the bottle now. Are we supposed to pretend they don't exist? Should the americans 70 years ago pretended jet engines weren't a thing because the Nazis invented them? I don't think so.
I see these questions often and they feel a little.. manipulative (perhaps unintentionally?). Because I've never seen anyone ask that we pretend LLMs don't exist. It's a straw man. Why do so many people bring it up? Where/how does this completely disjointed thought pattern originate?
Rather than pretending LLMs don't exist, I see demands to hold accountable the criminals who created them, as well as all the other criminals whose crimes are now being produced on an industrial scale with the help of said LLMs. Demands to limit how much further oppression can be carried out by the rogue companies, demands for fair compensation to those who spent their lives creating something that got exploited without consent, etc. Suggestions to perhaps not use these things indiscriminately, let alone pay for subscriptions until the companies start playing a fair game, etc.
Not intending to put words in JCAPER's mouth, but I read it as a turn of phrase, where "pretend they don't exist" == "not to use them". I expect that that thought pattern arises because, for...
Not intending to put words in JCAPER's mouth, but I read it as a turn of phrase, where "pretend they don't exist" == "not to use them". I expect that that thought pattern arises because, for example, a friend asked me to stop describing a tool another friend was using as "vibe coded" because just saying the word made them ill. Others have voiced that discussing any AI-related topics (e.g. on tooling I'd been working on to help me manage some focus and task management-related issues) in their vicinity makes them very uncomfortable, and that I should stop immediately.
My takeaway from such interactions is that people are pressuring me into not doing something, not expressing a nuanced agenda about investment and deployment of AI tooling. And as far as I can tell -- since the topic comes up extremely rarely as I always try to avoid it -- they don't have a clear set of goals like the ones you'd noted; they just want LLMs/generative AI tooling gone and for things to go back to normal.
I don't think your perspective and JCAPER's are mutually exclusive, though ...? It didn't seem like anything else they brought up contradicted your goals.
Thanks for explaining. I agree that there isn't much difference in our opinions. I'm just getting tired of seeing people parrot the "AI is here to stay, there's nothing we can do about it"...
Thanks for explaining.
I agree that there isn't much difference in our opinions. I'm just getting tired of seeing people parrot the "AI is here to stay, there's nothing we can do about it" black-and-white doomerism because in most cases it seems to imply that we have to take everything an AI company says or does as unavoidable. Use it or get left behind is another similar soundbite. And so on and so on.
It's as if Frodo went to Mordor and sent word back to Rivendell: "It's over, there are orcs everywhere and the air stinks. Nothing we can do, time to go home folks."
I wasn't intending to accuse @JCAPER in particular for this. His comment was well written and described many pain points accurately. I'm just inclined to call these propaganda-like statements out when I see them (when in an environment I actually care about).
Sorry to hear that some of your friends are reactive to the point of making you feel pressured. I can relate to their sentiment when so much of what I used to love about the world is being destroyed, mutilated, twisted into all sorts of sick forms against the original creator's will and intent, exploited, milked dry, what have you. The pain I feel can't be escaped because AI and the toxic people it enables (not you!) are everywhere. It's like a subtle form of torture and I can feel myself developing an allergy-like reaction to it. For example, nowadays just seeing the words "An honest question" feels extremely off-putting. As if I'd taken out a loaf of hand baked bread from its paper bag to discover it completely covered in mold.
Would it help at all to think of these reactions as potentially having nothing to do with you? If someone has food poisoning and ends up throwing up on your shoes, it's not necessarily because they're trying to influence you. They're just having a reaction that's out of their control. I know it isn't as cut and dried as this but maybe it can help alleviate some of the emotional weight of those experiences?
Yeah, that's definitely understandable, and I can relate. I personally find it very distasteful how these massive corporations are shaping the messaging. Re. my friends, it's a complex situation,...
Yeah, that's definitely understandable, and I can relate. I personally find it very distasteful how these massive corporations are shaping the messaging.
Re. my friends, it's a complex situation, but thank you. For what it's worth, I despise the way this whole situation has gone down in ways I lack the words to describe.
The last few months before departing what was to be the end of my software development career, I interrogated myself daily, inwardly considering "does this differentiate me from an AI? And if not, I should automate it away". Writing or reviewing code -- sure. Authoring architectural plans -- alright. Speaking to, or coordinating with teams. OK. Writing peer reviews, reading through applicants' resumes ...
What was the point of it all if there were trillions of dollars being spent on obsoleting me anyways. Why stay on the ride if it clearly ends in a brick wall.
I wanted to be a musician, at one point. I sold that dream for money. I wanted to write fiction, and inspire others to dream. I burned that desire to keep my heart warm in a career that sought to drown me. Friends and family were kept at arm's length, as the only thing I consistently felt was a dull anger. In the end, all I had left was the pride in my work, and I was instructed to sacrifice that too, on the altar of efficiency, to earn my manager a pat on the head (and a coin in their pocket).
I get that none of this is similar to what others are feeling, since if nothing else, experience has taught me that I think and feel in ways that're completely alien to the overwhelming majority of people. But, to relate to others' perspectives, I think back to how I felt when I reached inside to throw one more piece of my soul overboard to stay afloat, and grasped a clenched fist.
Anyways, I'm probably being overly dramatic (and I've had time to reflect and heal anyhow). Point is, I get that peoples' responses are coming from a place of hurt, and I try to give them as much space as possible until they feel comfortable again. It sucks a bit for me, but only insofar as I need to find other spaces to discuss AI topics (as noted elsewhere, the world is not lacking in those) and I need to add another conversation filter to the stack (it's already a mile high, so it's nbd).
I think we missed each other here. It's as @kacey said, what I meant by this is that there's a pressure on people who use LLM's by the "anti-AI crowd" (sorry if this sounds weird) to not use them...
I see these questions often and they feel a little.. manipulative (perhaps unintentionally?). Because I've never seen anyone ask that we pretend LLMs don't exist. It's a straw man. Why do so many people bring it up? Where/how does this completely disjointed thought pattern originate?
I think we missed each other here. It's as @kacey said, what I meant by this is that there's a pressure on people who use LLM's by the "anti-AI crowd" (sorry if this sounds weird) to not use them - or using my original metaphor, to pretend they don't exist.
I'm actually on board with a lot of what you said. I would love to see Scam Altman (not a typo) in prison, along with many others - which, sadly, considering who is in the US government now.... I'm not holding my breath sigh. And I have high hopes for chinese models. With each year they're producing ever smaller-but-capable LLMs that can run in consumer hardware - expensive hardware, granted, but still. My aspiration is to eventually self host my own LLMs and not pay a dime to Anthropic or others.
Thanks. I don't hate the idea of AI, I just hate the companies that are developing it for all the wrong reasons because those reasons shape the way it's being marketed to the public and the use...
Thanks. I don't hate the idea of AI, I just hate the companies that are developing it for all the wrong reasons because those reasons shape the way it's being marketed to the public and the use cases they focus on. And I do believe it's important to oppose them not just in spirit but in actions as well.
This is basically where I'm at, with the caveat that I will refuse to use any creative project if it's been made with AI, because that is art theft and I'm not ok with that as an artist. I think...
This is basically where I'm at, with the caveat that I will refuse to use any creative project if it's been made with AI, because that is art theft and I'm not ok with that as an artist.
I think AI has its place in the world, and I'm not against using it to make our lives easier. BUT I am against it being used to justify mass layoffs, doxxing, stealing data, the environmental impact LLMs and data centers have on the world, those who "created it" (used in quotations because we all know that those who are cool with theft of one kind of IP are cool with others) being the be-all, end-all of usage, how it's making people stop using their brains, and basically any other negative side effect of them that I just can't think of right now.
There are appropriate uses of AI/LLMs, and I'm fine with that. Using them to mess with call center scammers? Cool. Using them to help transcribe my doctor's notes and my conversation so something doesn't get missed in my chart? Sure. Do I want AI to have my PHI and use it to make my insurance premiums go up? Hell NO. Do I want creative projects like books or video games, or hell even flyers for my local commuity center events to be made with gen AI? absolutely not.
No I never cared about that. Specifically in the case of code generation I think it's not such a clear cut case. The models are trained on freely distributed copylefted code. Does that mean the...
Do we just not care that the tool itself is based on stolen work?
No I never cared about that. Specifically in the case of code generation I think it's not such a clear cut case. The models are trained on freely distributed copylefted code. Does that mean the resulting AI generated code should be copylefted as well? Maybe. I don't really care if a judge decides AI generated code is all AGPL now. I'd actually be ecstatic if I could make legally enforceable requests to every SaaS company around the world to hand over their now poisoned source code. I'm more than happy to have contributed my public code on GitHub to the model's training data. I've gotten more out than I put in. For image and video generation (which aren't tools I use anyway) there's a much stronger argument against it as so much more of the training data wasn't meant to be freely distributed and copied.
That continued use and support of it is actively destroying the environment for little benefit?
Fair point. And it's a shame because it's mainly an issue of electricity sourcing. It doesn't need to have its current carbon footprint. We should be running 100% renewable electrical grids by now. We are at least making good progress as a species on that. Solar is incredibly cheap, battery tech is improving. I personally wouldn't mind if datacenter expansions were limited by whether they could run off of clean energy.
That its growing commodity is causing a general decrease in individual knowledge and learning capacity?
Depends on the person. I'm able to do new things and learn in the process using these tools. I just finished a project which I will write up and share on Tildes soon. I used LLMs heavily throughout the process (for CAD, programming, data science). I could have done everything by myself, but it would have taken more hours to do so. It's not a kind of project I'd ever done before and I'm quite proud of it.
That it's making the industry of IT that most of the people commenting belong to worse to be a part of?
I'm outside of the software employment world at the moment. Thanks to LLMs I'm hoping to get into a niche area of automated manufacturing soon. I'm highly confident I can build my own CNC machine (not a CNC mill to be clear) given the availability of the basic components/modules/plans online combined with my knowledge and LLMs to fill in the gaps. It helps I have both an industrial warehouse and a makerspace to help me. One aspect I hope to see a tremendous uplift in is patching CNC firmware with LLMs to implement my customizations. Something that could have been a couple weeks of learning will probably be accomplished in a couple of hours. Sure I won't have a personal deep knowledge of the firmware - but I don't think I'll need to. When you build a more standard CNC device you usually don't need to know how the firmware works.
Edit: Honestly if you're someone with technical skills right now I think stretching your capabilities through LLM assistance is a goldmine. Things where you'd typically need to work on a small team (or just be skilled in many areas) can now be accomplished for nearly free as a solo entrepreneur. The LLMs from the top labs are currently subsidized. They only recently got good enough to use in this way. Now's the time to act!
I'm a software dev, and I should preface this with I've been on the rather anti-AI side. I think most of it comes from an inherent distrust of the companies providing it. I haven't ignored it...
I'm a software dev, and I should preface this with I've been on the rather anti-AI side. I think most of it comes from an inherent distrust of the companies providing it. I haven't ignored it completely. I've tried a few times, and found it something that I have to manage and work around more than a useful addition to my workflow.
Honestly if you're someone with technical skills right now I think stretching your capabilities through LLM assistance is a goldmine. [...] The LLMs from the top labs are currently subsidized. They only recently got good enough to use in this way. Now's the time to act!
I have thought about using it more for personal projects. I have some things I wouldn't mind spending more time on, but am usually too burnt at the end of the day/week to actually do it. But one of my biggest concerns is getting hooked and reliant on these tools to do anything, and then the subsidies finally end. They jack up the costs. Then what? You either pay up, or go back to not being able to use them. I sort of refuse to pay for the privilege of writing code. I don't know if I have a point or a question. Just speculation about capitalism constantly creeping into every facet of my life. It feels awful.
I started to feel comfortable using LLMs more heavily after using the open weight models that are ran profitably. Try GLM 5.3 through openrouter+opencode. Itās a fraction of the API cost of Opus 5...
I started to feel comfortable using LLMs more heavily after using the open weight models that are ran profitably. Try GLM 5.3 through openrouter+opencode. Itās a fraction of the API cost of Opus 5 and nearly as capable.
Edit: Consider this subscription if you want to stay away from the shady American LLM companies: https://opencode.ai/go
(seconding this -- IMO Qwen3.8 27b w/a 4-bit quantization even runs nicely locally on a B70 (32 GB Intel GPU). If you have a decent NVidia card with 12 GB+ of RAM, you could probably run one of...
(seconding this -- IMO Qwen3.8 27b w/a 4-bit quantization even runs nicely locally on a B70 (32 GB Intel GPU). If you have a decent NVidia card with 12 GB+ of RAM, you could probably run one of the older 35b MoE models as well. I have it in the other workspace autonomously troubleshooting a largely undocumented, obscure FOSS project. It's worked well for me so far, at least, but I haven't written anything substantial with it yet)
(I'm not OP) :fist_bump_emoji: samesies low key worried that someone cloned me now XD and yeah, been banking on this since late last year, so I'm stoked to see it coming to fruition. Hopefully...
(I'm not OP)
Thanks to LLMs I'm hoping to get into a niche area of automated manufacturing soon.
:fist_bump_emoji: samesies
[...] right now I think stretching your capabilities through LLM assistance right now is a goldmine. Things where you'd typically need to work on a small team (or just be skilled in many areas) can now be accomplished for nearly free as a solo entrepreneur.
low key worried that someone cloned me now XD and yeah, been banking on this since late last year, so I'm stoked to see it coming to fruition. Hopefully plans work out and I can meaningfully dig into this over the winter.
(edit) only thing I'd note is that local models have been consistently trailing frontier models by ~6-12 months in coding capability, at least, for a couple of years. So there's a possibility that losing those subsidies -- when the market corrects -- will barely be a speed bump. Fingers crossed ...
Iām planning to start in October. In the end Iāll have a machine that will likely have no equals around the world. Iāll be able to take orders online from a beautiful website with 3D previews and...
Iām planning to start in October. In the end Iāll have a machine that will likely have no equals around the world. Iāll be able to take orders online from a beautiful website with 3D previews and have something produce a real life equivalent. This will take something that normally would involve human labor and do it for cheap. I can lower the price dramatically, expanding the pool of customers. I plan to make multiple websites targeting different product types but all handled with the same machine.
I have actually made some of these widgets by hand and thereās demand. But I realized the prices Iād want to charge are too high. So automation is coming to the rescue.
That sounds really exciting! And yeah, super similar to what I'm aiming for as well (down to the labour price as a factor as well), so I selfishly hope you share some details about your process XD...
That sounds really exciting! And yeah, super similar to what I'm aiming for as well (down to the labour price as a factor as well), so I selfishly hope you share some details about your process XD I've always hoped that someone else would land on this train of thought, too, since I've felt quite mad being the only person in the room who's bringing it up.
Well thatās kind of the entire point of automation. I think Iāll also be able to achieve higher levels of consistency, quality, and with less time per widget. I havenāt done anything yet so I...
Well thatās kind of the entire point of automation. I think Iāll also be able to achieve higher levels of consistency, quality, and with less time per widget. I havenāt done anything yet so I canāt say exactly what Iāll do. But you can bet your ass Iāll be using a lot of 80/20 extrusion!
Mmhm! One of a couple leaps of faith from my perspective is (was?) whether the R&D for developing those automated manufacturing systems would be too time intensive/specialized, since that cost of...
Mmhm! One of a couple leaps of faith from my perspective is (was?) whether the R&D for developing those automated manufacturing systems would be too time intensive/specialized, since that cost of that needs to get amortized into your eventual product. Hence why I'm attempting to move somewhere with a much lower cost of living, in order to buy myself more swings at the bat so to speak š . Additionally, although manufacturing automation/higher throughput machine design is well established at the mid to high scale, it's unclear whether there are unforeseen challenges that would slow down development and deployment at the low end ... there's a whole spiel there that I'm wholly unqualified to talk about, but that I'll hopefully take a bite out of soon enough. Looking forward to 2027! XD
And good luck to you as well! If you're based in America, hopefully you can find a local source for your aluminum extrusion too š
The warehouse I mentioned is actually full of a ton of old 80/20 from a defunct robotics startup. So itāll be free! There are also a ton of cameras and optics there which were made for machine vision.
The warehouse I mentioned is actually full of a ton of old 80/20 from a defunct robotics startup. So itāll be free! There are also a ton of cameras and optics there which were made for machine vision.
Sorry if I'm being presumptuous but I think what you might mean here is that the denouncing of LLMs isn't as unanimous as you'd like. I'd say there's a pretty even mix of sentiment about AI on...
I'm saddened by how people on here seem to not only accept genAI programming so readily, but also completely dismiss ethics when justifying why they do so.
Sorry if I'm being presumptuous but I think what you might mean here is that the denouncing of LLMs isn't as unanimous as you'd like. I'd say there's a pretty even mix of sentiment about AI on Tildes, with hate edging out non-hate.
The second part of the quote is frustrating. It's blanket assigning a binary sentiment to a large group of people. All anyone can do with that is disagree or ignore it, it's too generalized to be true.
It's far more nuanced. For example, I think AI tools are incredibly useful for some things. Game changing. And I hate that a world changing technology is ultra-capitalized from the start. With the internet we had decades of exploration before the worst of the enshittification set in. That's unlikely to be the case with AI.
And the evironmental concerns are real (if often overexaggerated) and I care about that deeply, I think most people do. I also dislike the potential for AI to contribute to wealth inequality and consolidated control. However, like the internet, AI also increases individual capability. One person, or small groups of people, can accomplish much more than they could before. Sometimes more effectively than big organizations.
Here's the thing: It's not going to go away. It doesn't matter what you think, or I think. If we could somehow burn down all the big AI labs, people would advance the tech in their garages. A new round of funding would launch countless startups in the space. That's because the tech is genuinely useful. There was a lot of emotionally motivated writing about how LLMs don't really help anyone, and anyone who thinks they do is just deluding themselves, in 2024 and even early 2025 but that's painfully outdated. It's only people with no expertise in any of the fields where AI has value that are saying that kind of thing now.
I have nothing against a website listing AI free options. I think it's great, bring on the demand for artisanal, entirely hand crafted, options.
But also the reality is that, in the field of software engineering at large, AI assisted (as opposed to wholly vibecoded) development is already the rule rather than the exception and nothing is going to change that.
You're right about this and I apologize. I've been too angry about the topic lately, for a number of reasons, and I let that emotion take the head of my discourse. It's not in the spirit of Tildes...
Sorry if I'm being presumptuous but I think what you might mean here is that the denouncing of LLMs isn't as unanimous as you'd like. I'd say there's a pretty even mix of sentiment about AI on Tildes, with hate edging out non-hate.
The second part of the quote is frustrating. It's blanket assigning a binary sentiment to a large group of people. All anyone can do with that is disagree or ignore it, it's too generalized to be true.
You're right about this and I apologize. I've been too angry about the topic lately, for a number of reasons, and I let that emotion take the head of my discourse. It's not in the spirit of Tildes and could have been seen as uncivil, so while I don't agree with the responses, I appreciate that they were more level-headed than mine.
I do stand by the fact that I can't support AI-assisted development (or genAI in general), regardless of how commonplace it's become. I do think that, in a different form, I would at least be able to accept it. My problem is with OpenAI, Anthropic and others relying on or producing and maintaining behemoth models in ways that have no business being tolerated, but I've had to remember we were already studying what we know as AI before 2022 and there's people using the tech to do fascinating and useful things without relying on scraped data and misuse of energy (even for art! I learned how SynthetizerV and similar products do their work recently and it's a great example). The concept itself isn't immoral to me.
There might be a day where it's possible and viable to get assistance from a local SLM trained only on data that was fairly obtained from consenting authors (I think it's even likely). That, I likely still wouldn't use unless necessary, because I've been down the road of using LLMs for programming and found it hurt my competency more than it helped; but I would at least be okay with its existence.
Thanks for the civil response! No argument with you choosing to avoid LLMs. Even in software engineering I suspect there will be room for competent people who don't want to use agents for... well...
Thanks for the civil response! No argument with you choosing to avoid LLMs. Even in software engineering I suspect there will be room for competent people who don't want to use agents for... well I don't know how long, but maybe quite a long time.
A lot of these issues feel like they're because we're in the transition from the "legacy" world into the post-AI world. During the transition, everything feels broken. There's an adaptation. It's...
A lot of these issues feel like they're because we're in the transition from the "legacy" world into the post-AI world. During the transition, everything feels broken. There's an adaptation. It's pretty obvious with the current pass that's happening in cyber security.
In terms of stolen work, since I've started learning about Linux and free software, I've always felt like the current system was broken. https://web.mit.edu/m-i-t/forums/copyright/index_transcript.html The culture change that AIs bring seem to me to be something that will be positive.
For the environment impact, I guess I got magical thinking, but I feel like it's the only way to speed up research to save the earth.
In terms of benefit, I'd disagree, the company I work for, for example, uses it to save lives.
I don't think my response is that good, but I didn't have time to think it through fully. Will probably revisit it later. These are my general ideas.
LLMs as they are right now (opaque closed-source models, tightly controlled and guarded by billionaire-owned corporations) are the complete antithesis to FOSS ideals. They are an extreme shift in...
In terms of stolen work, since I've started learning about Linux and free software, I've always felt like the current system was broken. https://web.mit.edu/m-i-t/forums/copyright/index_transcript.html The culture change that AIs bring seem to me to be something that will be positive.
LLMs as they are right now (opaque closed-source models, tightly controlled and guarded by billionaire-owned corporations) are the complete antithesis to FOSS ideals. They are an extreme shift in the opposite direction of what you're implying here.
For the environment impact, I guess I got magical thinking, but I feel like it's the only way to speed up research to save the earth.
You are correct, this is 1000% magical thinking. Even if it were feasible that predictive text generating engines could somehow come up with a solution to our climate problems, none of these companies are currently or will ever bother actually putting any kind of meaningful resources behind that kind of research, because it won't increase their profits (and I mean, realistically they probably know that the idea is absurd to begin with).
The thing about the frontier models is that theres no real secret sauce. They advance mostly based on throwing raw compute power into training, which is why all of the frontier labs have similar...
LLMs as they are right now (opaque closed-source models, tightly controlled and guarded by billionaire-owned corporations) are the complete antithesis to FOSS ideals
The thing about the frontier models is that theres no real secret sauce. They advance mostly based on throwing raw compute power into training, which is why all of the frontier labs have similar performance. Those models get distilled, and cheaper Chinese and open source models see similar performance within six months to a year. Openai, anthropic and Google are controlled by billionaire run corporations, LLMs as a whole aren't though.
One of two things is going to happen. We're going to see increasingly diminishing returns, where in order to get any improvement, more capital and power than exists on the planet is required, and open source competitors will close the gap and catch up. Or, these models will never stop getting significantly better, in which case we're likely fucked as the technology dominates the human race.
I think the former is the more realistic option, but I don't see anything besides harsh multilateral government enforcement of a total ban being feasible if it's the latter.
The current versions of LLMs are opaque, but there's a consensus that we're in an AI bubble which could pop at any time. The open models will be run on cheaper and cheaper hardware overtime. But I...
The current versions of LLMs are opaque, but there's a consensus that we're in an AI bubble which could pop at any time. The open models will be run on cheaper and cheaper hardware overtime. But I didn't mean it that way only, there's a part about how it's a good tool for compacting knowledge. Also last time I checked, the output from LLMs couldn't be copyrighted.
It's not the predictive text that would necessarily discover something of value (though in math they are making some pretty significant breakthroughs without much guidance), but its ability to speed up research and how researchers use it as a tool.
If I were you I'd be concerned it's a way to subconsciously bury my head in the sand, as there isn't any research required (which, to be clear, isn't the same as saying we shouldn't do any)....
For the environment impact, I guess I got magical thinking, but I feel like it's the only way to speed up research to save the earth.
If I were you I'd be concerned it's a way to subconsciously bury my head in the sand, as there isn't any research required (which, to be clear, isn't the same as saying we shouldn't do any). Perhaps the biggest impediment to environmental progress is the lack of any consensus that we should make progress. In this I'd say AI is on balance an impediment, as well as being environmentally detrimental directly.
It is the current system that affords authors the right to license their work under whatever terms they please. There is no Free software, with its strict licensing requirements, without a legal...
In terms of stolen work, since I've started learning about Linux and free software, I've always felt like the current system was broken. https://web.mit.edu/m-i-t/forums/copyright/index_transcript.html The culture change that AIs bring seem to me to be something that will be positive.
It is the current system that affords authors the right to license their work under whatever terms they please. There is no Free software, with its strict licensing requirements, without a legal system that protects these rights.
You'll be less pleased to know that one frequent use of LLMs is to perform so-called copyright laundering of Free software. Modules, libraries or whole applications are fed into the LLM which rewrites them ever so slightly so as to rid the software of signs of the original authorship and, at least ostensibly, the licensing requirements imposed by the original author. The idea is that by doing this you can more easily create proprietary software based on Free software.
The negative effects of LLM use on the maintenance of Free and Open Source software are well known. For every useful contribution to a popular project there are a hundred patches and PRs by useless charlatans who practically function as transparent proxies for the LLMs they use by uncritically reproducing their hallucinations, neither understanding the problem they may or may not be solving which may or may not exist nor understanding the solution they submit. This has made many F/OSS project maintainers more restrictive in what kinds of contributions they'll accept or entertain at all, potentially throwing some babies out with that dirty, smelly ocean of bathwater.
For the environment impact, I guess I got magical thinking, but I feel like it's the only way to speed up research to save the earth.
We know what the problem is and we know what we are doing to cause that problem. The reluctance to solve the problem accordingly is a political phenomenon.
Except that free software only exists to counter the current system. There wouldn't be a need for free software licenses if we didn't have to fight fire with fire. You can read more about it in...
I agree that LLMs can be used to launder free software. My idea above was for use in the opposite direction. You can use an LLM to recreate currently closed source programs. (Last week I got my own implementation of a software that's closed and requires admin rights and runs permanently as a background service. Now that I have the code, I'm testing to make sure that it works well before I release the code.) So I'd say you can also create more software in general and it works both ways.
While the negative effects are well known in open source projects, I see them as being part of the transition from our legacy world. It's allowing more people to contribute even though they don't have the knowledge to. That has negatives, but we'll learn to deal with it and people will get better. I think Linus Torvalds' stance on this is the right one.
Knowing the problems on environnement and being able to fix them is completely different. The only thing that seemed to work so far was during the pandemic lockdowns. Most people don't want to go through that again. We're already locked into bad outcomes and predictions are happening faster than expected. Vicious cycles are going to get worse and worse. I don't expect anything to get better the way things are going.
It's not so much a matter of opinion on what LLMs can and can't do as an observation of an ongoing phenomenon. Yes, plagiarism is to be expected everywhere, not just of Free software. What does...
I agree that LLMs can be used to launder free software.
It's not so much a matter of opinion on what LLMs can and can't do as an observation of an ongoing phenomenon.
You can use an LLM to recreate currently closed source programs.
Yes, plagiarism is to be expected everywhere, not just of Free software.
I see them as being part of the transition from our legacy world.
What does this even mean? What is a "legacy world" and what are we transitioning into? Will the world you think we're transitioning into not have a legacy? I try to think clearly about this issue, but terminology like this isn't making things easier. It seems to me like a creative alternative to waving your hands instead of presenting an argument.
Knowing the problems on environnement and being able to fix them is completely different. The only thing that seemed to work so far was during the pandemic lockdowns. Most people don't want to go through that again. We're already locked into bad outcomes and predictions are happening faster than expected. Vicious cycles are going to get worse and worse. I don't expect anything to get better the way things are going.
With that perspective in mind, the hope in LLMs magically solving the problem for us makes even less sense.
Legacy to me is how there are many issues with the world right now. A lot of those issues are invisible. You can look at cybersecurity, the ways AIs can get into systems already existed. The bugs...
Legacy to me is how there are many issues with the world right now. A lot of those issues are invisible. You can look at cybersecurity, the ways AIs can get into systems already existed. The bugs didn't appear because of AIs, they only get exposed by them. The speed that they get exposed increases as AIs improve. I say legacy because at some point, there should be fewer and fewer issues to patch. When that happens, that's when the transition is done. While it's happening, it looks like it's the end of the world for cybersecurity. But once you're on the other side, the world should be better since there are fewer bugs.
The same thing can happen with everything else. When you can't trust the news online, perhaps we go back to local news that can be verified easier.
It think it's exposing issues that were already there and that we have to deal with.
On plagiarism, I see it more as unlocking knowledge to the whole world which is a good thing. Calling it plagiarism implies that only one project of any kind is allowed to exist at once.
That "when" assumes a particular answer to a very big "if". I'm seeing no indication yet that LLMs will ever be used to solve bugs faster than they are used to produce them. The economic...
When that happens, that's when the transition is done.
That "when" assumes a particular answer to a very big "if". I'm seeing no indication yet that LLMs will ever be used to solve bugs faster than they are used to produce them. The economic incentives are pointing elsewhere: access to models that can thoroughly audit and produce high quality threat analyses that identify critical vulnerabilities is much more expensive than access to models that can produce poor to mediocre code which ostensibly implements solutions to business domain problems but has an error rate similar to that of a junior developer.
There are a few differences between the LLM and the junior developer that further exacerbate the problem. The junior (without an LLM) is limited in their productive capacity by their lack of skill, knowledge and wisdom. They will produce code at a lower rate than the experienced developer, because they will have to stop to learn things more often, ask more questions or will simply avoid problems that seem challenging. The burden imposed on more skilled team members in reviewing and auditing the low-quality code is consequently limited. The LLM knows no such limits and will spit out massive changes without ever gaining knowledge. I think people working in engineering organizations that are being pushy about LLMs can corroborate my experience that this eventually results in lower quality reviews.
I worked in an organization where management "addressed" the problem by reminding developers that their code is their responsibility, whether they wrote or generated. That sounds nice in theory, but is ultimately simply not true. The resulting problems won't magically be directed at the person who created them, but at the team that maintains the software.
You should consider that what works for Linus Torvalds might not work for every other software project. Linux has generally high standards of inclusion because it deals with the production and maintenance of software responsibly and uncompromisingly as a high-stakes engineering problem. Being as widely deployed as Linux is also attracts a lot of attention from security researchers and skilled developers, funded by organizations with legitimate interests in its quality. I am however using (and I'm being used in) a lot more software produced by organizations with much lower standards, driven by the market to "get shit done", "move fast and break things" or whatever other slogans they use to justify being quick to market at the cost of quality of engineering. This use of these systems is often not voluntary.
Meanwhile, with the advent of LLMs, the difference in cost between producing such low quality systems and producing high quality systems has increased. I think it will increase further as investors become impatient waiting for the chatbot vendors to realize the potential they see in them as profitable businesses. If any charlatan with no knowledge of engineering can use an LLM to quickly generate volumes' worth of low-quality software alone and something with the rigor of Linux still demands the attention of hundreds of engineers on the LKML, the production of software will be much more in the hands of incompetent charlatans than people like the Linux maintainers.
Itās a bit ironic to complain about ācopyright launderingā when free software got it start by reimplementing proprietary software. Stallman wrote copylefted code that reimplemented proprietary...
Itās a bit ironic to complain about ācopyright launderingā when free software got it start by reimplementing proprietary software. Stallman wrote copylefted code that reimplemented proprietary Unix commands that did the same thing. This was done specifically to evade the licensing requirements of the original software!
And to be clear, I think implementing free alternatives to proprietary software is a good thing, though the software vendor might not like how it undermines their business model. If youāre trying to make a living writing software, competition from free alternatives is a problem. (Free only after theyāre implemented of course. Actually building the software still costs time and money, though less than before.)
Sometimes people hoped that free software would become āviralā due to network effects along with copyleft restrictions. But to the extent that it has, itās mostly not due to the copyright restrictions. Every popular programming language has a huge collection of open source libraries available and copyleft doesnāt play much of a role in that. And now that AI-driven rewrites are feasible, itās even more dead. Itās too easy to make clones that are free of whatever copyright restrictions you might add.
The remaining āmoatā is basically trust, along with inertia. Just because some unknown author published an AI rewrite and it seems to work, would you trust it? How exactly do new projects become trusted? We often depend on network effects - if an open source library has been around for a while and seems to be used a lot, maybe itās okay? We donāt even know the names of most of the developers whose work we depend on.
When a new version is published under the same name, hopefully upgrading is safe? Nowadays package managers are adding ācooldownā periods so that ordinary developers donāt upgrade immediately. There are organizations that automatically scan new releases for malware, and adding a delay gives them time to sound the alarm. I wonder to what extent these vendors are starting to use AI for their scans?
As a wee note, afaik the only research area of note is direct air carbon capture. Everything else seems addressable with current day technology, but it runs into social problems (people despise...
For the environment impact, I guess I got magical thinking, but I feel like it's the only way to speed up research to save the earth.
As a wee note, afaik the only research area of note is direct air carbon capture. Everything else seems addressable with current day technology, but it runs into social problems (people despise change) and class problems (currently wealthy folks operating companies dislike the prospect of uncertainty or reduced profits). R&D expenditure seems minimal beyond stuff like advancing fusion energy research, but that is slow because it takes a long time to build reactors and validate theory afaik.
Thereās also various forms of geoengineering (mostly in the area of solar radiation management), but those are a lot more controversial. As far as direct air capture, thereās not really technical...
Thereās also various forms of geoengineering (mostly in the area of solar radiation management), but those are a lot more controversial. As far as direct air capture, thereās not really technical limitations. Thereās already pilot plants running. The real issue is cost/efficiency (they are massive energy hogs), which might become moot with either cheaper overbuilt renewables or micro-reactors (especially if they can cogen and provide process heat).
Afaik, the primary concern is that DAC plants constantly run into fouling problems with their catalysts. An AI could -- hypothetically -- scan through potential alternative catalysts which might...
As far as direct air capture, thereās not really technical limitations.
Afaik, the primary concern is that DAC plants constantly run into fouling problems with their catalysts. An AI could -- hypothetically -- scan through potential alternative catalysts which might be more durable, or otherwise automate the testing process at small scales (similar to how we do microfluidic assays) in order to rapidly assess viability for groundbreaking approaches. Since right now the problem with direct air capture is that it'd need, like, some absurd double digit percentage of the world's power generation to make a big enough dent to be viable (as you note, but hence the need for advancement). In other words, overbuilt would have to look like blanketing the Sahara in solar panels or a breakthrough in fusion reactors.
Er, all of which is to say, what're currently referred to as DAC technologies are unlikely to advance much, afaik (circa like ~2022 which is the last time I did any reading on this). (edit) (for context, from my non-expert, limited research perspective, wood vaulting or some twist on ocean fertilization are the most promising technologies we have for carbon sequestration)
Furthermore, w.r.t. solar radiation management, I think we already know as much as we can without experimentation around stratospheric sulphur aerosol injection which afaik is the only practical method anyone's proposed. Fingers crossed that space mirrors work out, but I'm not banking on it ...
You're getting a storm of pro-AI responses from the looks of things, so I just wanted to let you know I really appreciate your post. I've long felt that any time I enter a conversation with...
You're getting a storm of pro-AI responses from the looks of things, so I just wanted to let you know I really appreciate your post. I've long felt that any time I enter a conversation with someone pro-AI here that it feels like I'm talking to a brick wall that is determined never to accept any of the arguments against AI usage, and it drives me nuts.
From my perspective, it feels like an automatic defensive measure; people got interested in the tech for one reason or another, and now that they've gotten used to it they don't want to switch back. So they find the reasons they can to justify it against the people telling them to stop, however poor they may be. If this is an accurate assessment, I can still have hope that models will eventually force them to re-evaluate when prices inevitably jump, but damn if it isn't frustrating.
Just want to provide the info that there are models served at a profit right now from Chinese vendors and theyāre more than good enough for professionals to use for building software.
I can still have hope that models will eventually force them to re-evaluate when prices inevitably jump
Just want to provide the info that there are models served at a profit right now from Chinese vendors and theyāre more than good enough for professionals to use for building software.
Trust meā¢ļø? Anyway, not for me. I don't have an issue with AI assisted coding. Vibe coding though, yes. Different story. I avoid those projects by default.
Tired of AI slop? clankerfree is a resource that indexes projects that the authors self-certify as not containing or utilizing generative AI or LLM technology.
Trust meā¢ļø?
Anyway, not for me. I don't have an issue with AI assisted coding.
Vibe coding though, yes. Different story. I avoid those projects by default.
I'm not who you asked but here's mine: AI assisted: I designed the code, with or without iterating in a solution with an LLM or another human. I may have accepted "autocompletions" provided by an...
I'm not who you asked but here's mine:
AI assisted:
I designed the code, with or without iterating in a solution with an LLM or another human. I may have accepted "autocompletions" provided by an LLM, or used a tool like Claude Code to describe a function and accepted a whatever it generated (after reviewing it and understanding how it works). I may even so large scale refactors using the same tools, but all under my review and guidance.
Vibe coded:
I described an end goal and let an LLM agentically / iteratively design and write it autonomously with minimal review or understanding of the underlying code.
The former is using tools to speed up my own workflow.
The latter is outsourcing entirely and blindly trusting something works as intended and is of good quality.
The way I see it is: it's the exact same as Google or StackOverflow. We don't say software is "Google assisted" because you searched online to find an answer. LLMs are the same, but much more...
The way I see it is: it's the exact same as Google or StackOverflow. We don't say software is "Google assisted" because you searched online to find an answer. LLMs are the same, but much more specific and faster. It can parse out documentation in minutes that would normally take hours.
Iāll add another thought: my coworkers with software development backgrounds using AI to help them would be AI-assisted coding. Me (with a very poor-understanding of coding and programming) using...
Iāll add another thought: my coworkers with software development backgrounds using AI to help them would be AI-assisted coding. Me (with a very poor-understanding of coding and programming) using AI to spit out a program would be vibe coding.
Purity test, if you ask me[1]. Yes, of course, Vibe coding is bad, nobody's contesting that, but there's nothing wrong with using AI in your coding projects provided you could conceivably do the...
Purity test, if you ask me[1]. Yes, of course, Vibe coding is bad, nobody's contesting that, but there's nothing wrong with using AI in your coding projects provided you could conceivably do the same work yourself, given enough time. To me, Linus Torvalds was correct on this issue.
I really feel like these project lists, but especially Codeberg as a whole, wants to desperately feel better than everyone else. And you know what? Fine, they're better. Gold star. You get to go to heaven because you didn't use it. There you go. Congrats. But provided the project is good and the operator knows what they're doing, there's no substantive difference between good hand-written code and AI-assisted code.
[1]: Also evidenced by the fact that to get on this list, you need to post on Mastodon about your project. What, no Mastodon? No ActivityPub? Well, then you can't be serious about open source software, clearly.
I think a "thought terminating cliche" while not the term I would use, is very different from "an attack on my person" and anyone who agrees with you. I think accusing someone of attacking you is...
I think a "thought terminating cliche" while not the term I would use, is very different from "an attack on my person" and anyone who agrees with you.
I think accusing someone of attacking you is serious enough to make this point despite it being tangential to the actual conversation. It certainly doesn't bring the temperature of any discussion down. And if your cynicism is impacting your experience so much as to accuse a link to be a personal attack, then I would suggest disengaging, even if only temporarily.
You talk about this like it's some infallible error-finder, and not something prone to contradicting itself each time it's run. You also talk about it as though doing this has no harms to weigh...
You talk about this like it's some infallible error-finder, and not something prone to contradicting itself each time it's run. You also talk about it as though doing this has no harms to weigh against.
I'm not getting into a debate with you on the utility of AI. I'm pretty sure we've done that before and it became abundantly clear you were not going to change your stance, and I don't think I've...
I'm not getting into a debate with you on the utility of AI. I'm pretty sure we've done that before and it became abundantly clear you were not going to change your stance, and I don't think I've ever seen you do anything but defend AI besides.
I'm just pointing out why the confidence and snark in your initial comment was undue.
Well, finding real vulnerabilities with some double digit probability for pennies, that others now can also discover for the same price doesn't sound like a bad deal. Certainly cheaper than...
Well, finding real vulnerabilities with some double digit probability for pennies, that others now can also discover for the same price doesn't sound like a bad deal. Certainly cheaper than getting all your data leaked/hacked.
I wonder how these projects would stand up to the watermarks that are being applied. Itās one thing to say, and possibly believe, that your project doesnāt have any LLM code in it, but if there...
I wonder how these projects would stand up to the watermarks that are being applied. Itās one thing to say, and possibly believe, that your project doesnāt have any LLM code in it, but if there are unknown outside contributorsā¦ and of course the library ecosystem is riddled with LLM code, so if you vendor any of those libraries you may unwittingly add it in yourself. Itās like microplastics these days.
Cute but the whole Git platform is vibe coded at this point I donāt know how anyone would actually avoid all vibe coded platforms. Also, the software to tell teachers if a paper is LLM generated...
Cute but the whole Git platform is vibe coded at this point I donāt know how anyone would actually avoid all vibe coded platforms.
Also, the software to tell teachers if a paper is LLM generated kinda sucks, I donāt see how this could be better.
Ive been collecting used books because anything published after about 2023 was at least edited by an LLM, if not written by one, and theres no way to tell for sure other than just trusting everyone involved.
All of them, github, bitbucket, gitlab, every single service that hosts git repos is vibe coded now. So if youāre trying to use Git, youād have to set up your own remote if you wanted to avoid...
All of them, github, bitbucket, gitlab, every single service that hosts git repos is vibe coded now.
So if youāre trying to use Git, youād have to set up your own remote if you wanted to avoid vibe coded projects.
On top of that:
Microsoft services are vibe coded, Google services, Apple services, your banking apps and websites, Doordash, Uber, Lyft, all online ordering apps, all social media, all entertainment services, all that shit has been vibe coded for the past two years.
I know the point is probably so that when youāre building a project you can avoid using other peoples vibe coded libraries because they may be more prone to breaking, but if youāre gonna do that you gotta host your own Git remote and I donāt even know what flavor of Linux has no vibe coded pieces in it. Everything has this shit in it.
Minimum remote for git is a directory, possibly accessible via ssh. ie: git+ssh://192.168.1.10/path/to/repo.git A remote can even be a directory on the same machine. I did both of these recently...
Minimum remote for git is a directory, possibly accessible via ssh.
ie: git+ssh://192.168.1.10/path/to/repo.git
A remote can even be a directory on the same machine. I did both of these recently to easily move code between my IDE projects dirs to something's plugins dir for a couple of things I was working on.
Git doesn't require any third party code hosting services to operate or exchange source code.
These services provide a (hopefully reliable) storage location and a web ui for the managing access & review, among other things.
As much as I like Tildes and understand the premise that we should all be trying to consider each others' opinions, I'm saddened by how people on here seem to not only accept genAI programming so readily, but also completely dismiss ethics when justifying why they do so.
There's a few people on here implying this would only be relevant if it only excluded vibe-coded projects, not AI assistance. That only the resulting product matters. Did we forget everything we were criticizing large models for in the first place? Do we just not care that the tool itself is based on stolen work? That continued use and support of it is actively destroying the environment for little benefit? That its growing commodity is causing a general decrease in individual knowledge and learning capacity? That it's making the industry of IT that most of the people commenting belong to worse to be a part of? Like, surely, a bit more than just the result matters here?
Ethics matter. Call it a purity contest or virtue signaling or whatever, I don't care. Trying to do something good to push against the bullshit even if it's imperfect and still has to rely on tech that does use it to exist and be effective is still better than doing nothing at all. This is a good thing and I have no criticism to make of it.
Yo š I'm personally trying to tone down how much confrontation I cause in this space, but I wanted to minimally note my disapproval. Framing this as a black and white ethical issue equally saddens me, since that same absence of nuance and appeal to morality has been used to justify atrocious actions throughout history. And it erases any good that the tools can accomplish. Furthermore, given the amount of personal attacks I get whenever I mention LLMs/AI here, I sincerely doubt that there's a majority of people who agree with me -- you are almost certainly in the majority. I'd imagine people who hate this have
artificial intelligencefiltered out of their tags, or otherwise avoid the topics.(edit: sigh ... I had hoped this wouldn't be exemplary'd. For anyone reading this, fwiw, the person was pleased with the notion of trying to keep conversations calmer. It wasn't an AI thing)
I was going to give an exemplary label because of this, but I wanted to show publicly why I 100% agree with your point.
I'm so, so, so tired of this 0 or 1 attitude. You're either 100% in favour of something, or 100% against it. It feels like nowadays on the internet you can't stand somewhere between both. This attitude I feel is more destructive in the long run.
I, for example, think:
I have a lot more thoughts, but these zigzag of thoughts should be good enough to illustrate my point. No one will ever convince me that these things are the best thing ever - we shouldn't underestimate some of the societal dangers these could potentially have. But neither will no one ever convince me that my own experiences are wrong, and that LLMs didn't have some good outcomes for my own skills and work.
I haven't posted on here a lot about LLM stuff, but I keep reading most of it...or at least, most of the threads on it, which have not been great for me mentally.
I think I'm fairly anti-ai, but I have a hard time really examining my position because I know how emotionally biased I am in it.
I am not an artist. I have never been very "creative", but I love the thought of being creative.
I loved legos growing up, and I think most of that is because it let me put things together
in ways that were creative and showing creativity, but did not require me to be "good" at it.
I was never able to get into drawing or anything like that in the same way, and I've always
loved reading comics and seeing what artistic people are able to create.
I went to school for programming, because I truly love writing code & seeing things happen as a result.
To me, programming has been my "creative" output. So I find myself biased against LLMs generating code, for what I imagine are the same emotional reasons as a lot of artists, but not necessarily the reasons that get sympathy or could actually have legal repercussions (e.g. training on stolen art).
But also I struggle with it myself, becuase like, am I gatekeeping? Am I against it because it was my "special thing" that many others couldn't do? I don't think so, but maybe there's some of that to it as well, as to why I don't like it. But like, at the same time, I'm glad to hear that it has helped you be able to set up a homelab and troubleshoot stuff on Linux.
I think it is great that people are now able to customize software and make tools and stuff that are useful to them, that they would never have been able to "before LLMs". Except, if I look at this from the art angle, I'm not sure how consistent my view is here?
I've never really been good at putting my feelings into words, and I certainly haven't gotten any better at it. It just seems like most of what I have been good at, are things that now people are using LLMs for.
and I don't know where that leaves me.
I'm going to keep avoiding LLMs for the things I enjoy doing, which is...most of the things I do,
and it does frustrate me seeing LLMs continually invade those things, and it is sometimes saddening
when I see others using LLMs for the things I enjoy doing. I don't think others should necessarily
not use them for those things, because if they don't enjoy doing them, sure, automate it away.
But like, it is much, much too late to put the genie back into the bottle, so all I can really do is personally protest by avoiding using them where I can. And I don't know whether I'm doing that for ethical or emotional reasons, but I'm also not really sure it matters.
I do really appreciate your take here, as well as from @kacey. I like being able to have some of these nuanced conversations around LLM uses, (or at least, read other peoples nuanced takes when I don't feel like commenting).
Yeah, I think I'm pretty aligned with you on my feelings around this stuff. I don't mind when other people use AI to help them do personal projects, or when people use it at work to help with work, as long as the quality is still good (big qualifier, I know).
I do absolutely HATE when people diminish the work of other people by commenting that AI could do that, or that it wasn't actually that hard because they did the same thing with AI. It fills me with blinding rage, and it's very common amongst obnoxiously pro ai spaces.
When someone hand builds a complicated piece of software, shares it with the world, talks about the years they've spent on it, and someone comments that AI could do the same thing, I just want to slap them.
Even if it was the same quality, which is extremely dubious, imagine someone going up to a master blacksmith that created a hand forged masterpiece of a sword over months and going "I could get the same thing on temu for 10 dollars".
This unfortunately happens (very common in custom furniture especially), but I absolutely detest the idea.
Even if the results were the same (they basically never are). Someone hand crafted by a human being is special, and will always be more valuable than something mass produced and shitted out by a machine. Even if blacksmithing and furniture making is no longer a viable career in most places, the people those things create has way more intrinsic value to most people, and the same is true for software in my opinion.
Thank you very much for not doing that š I feel that it needlessly escalates tensions on a topic that needs to be discussed (as noted in your comment!).
For my part, I've been wishing for years that the temperature would come down far enough to talk about effective resistance against all this -- job loss via automation, devaluing of the arts, legislative reform, etc. But that requires acknowledging that the problem exists, and dispassionately analyzing the space for asymmetries to be exploited in favour of the common people.
We need to be organizing workers, pressuring local governments, and pushing our vision of what a future with these tools should look like -- not de-facto rolling over and allowing OpenAI/Anthropic/Microsoft/Google to decide for us. Endlessly debating about data centres and morality cannot un-create the idea of a transformer, or delete every open weight model on the internet, but I firmly believe that the window of opportunity remains open for deciding whether the world on the other side of all this is a good one. For whatever definition we -- not Sam Altman or Dario Amodei -- decide "good" is.
I see these questions often and they feel a little.. manipulative (perhaps unintentionally?). Because I've never seen anyone ask that we pretend LLMs don't exist. It's a straw man. Why do so many people bring it up? Where/how does this completely disjointed thought pattern originate?
Rather than pretending LLMs don't exist, I see demands to hold accountable the criminals who created them, as well as all the other criminals whose crimes are now being produced on an industrial scale with the help of said LLMs. Demands to limit how much further oppression can be carried out by the rogue companies, demands for fair compensation to those who spent their lives creating something that got exploited without consent, etc. Suggestions to perhaps not use these things indiscriminately, let alone pay for subscriptions until the companies start playing a fair game, etc.
Not intending to put words in JCAPER's mouth, but I read it as a turn of phrase, where "pretend they don't exist" == "not to use them". I expect that that thought pattern arises because, for example, a friend asked me to stop describing a tool another friend was using as "vibe coded" because just saying the word made them ill. Others have voiced that discussing any AI-related topics (e.g. on tooling I'd been working on to help me manage some focus and task management-related issues) in their vicinity makes them very uncomfortable, and that I should stop immediately.
My takeaway from such interactions is that people are pressuring me into not doing something, not expressing a nuanced agenda about investment and deployment of AI tooling. And as far as I can tell -- since the topic comes up extremely rarely as I always try to avoid it -- they don't have a clear set of goals like the ones you'd noted; they just want LLMs/generative AI tooling gone and for things to go back to normal.
I don't think your perspective and JCAPER's are mutually exclusive, though ...? It didn't seem like anything else they brought up contradicted your goals.
Thanks for explaining.
I agree that there isn't much difference in our opinions. I'm just getting tired of seeing people parrot the "AI is here to stay, there's nothing we can do about it" black-and-white doomerism because in most cases it seems to imply that we have to take everything an AI company says or does as unavoidable. Use it or get left behind is another similar soundbite. And so on and so on.
It's as if Frodo went to Mordor and sent word back to Rivendell: "It's over, there are orcs everywhere and the air stinks. Nothing we can do, time to go home folks."
I wasn't intending to accuse @JCAPER in particular for this. His comment was well written and described many pain points accurately. I'm just inclined to call these propaganda-like statements out when I see them (when in an environment I actually care about).
Sorry to hear that some of your friends are reactive to the point of making you feel pressured. I can relate to their sentiment when so much of what I used to love about the world is being destroyed, mutilated, twisted into all sorts of sick forms against the original creator's will and intent, exploited, milked dry, what have you. The pain I feel can't be escaped because AI and the toxic people it enables (not you!) are everywhere. It's like a subtle form of torture and I can feel myself developing an allergy-like reaction to it. For example, nowadays just seeing the words "An honest question" feels extremely off-putting. As if I'd taken out a loaf of hand baked bread from its paper bag to discover it completely covered in mold.
Would it help at all to think of these reactions as potentially having nothing to do with you? If someone has food poisoning and ends up throwing up on your shoes, it's not necessarily because they're trying to influence you. They're just having a reaction that's out of their control. I know it isn't as cut and dried as this but maybe it can help alleviate some of the emotional weight of those experiences?
Yeah, that's definitely understandable, and I can relate. I personally find it very distasteful how these massive corporations are shaping the messaging.
Re. my friends, it's a complex situation, but thank you. For what it's worth, I despise the way this whole situation has gone down in ways I lack the words to describe.
The last few months before departing what was to be the end of my software development career, I interrogated myself daily, inwardly considering "does this differentiate me from an AI? And if not, I should automate it away". Writing or reviewing code -- sure. Authoring architectural plans -- alright. Speaking to, or coordinating with teams. OK. Writing peer reviews, reading through applicants' resumes ...
What was the point of it all if there were trillions of dollars being spent on obsoleting me anyways. Why stay on the ride if it clearly ends in a brick wall.
I wanted to be a musician, at one point. I sold that dream for money. I wanted to write fiction, and inspire others to dream. I burned that desire to keep my heart warm in a career that sought to drown me. Friends and family were kept at arm's length, as the only thing I consistently felt was a dull anger. In the end, all I had left was the pride in my work, and I was instructed to sacrifice that too, on the altar of efficiency, to earn my manager a pat on the head (and a coin in their pocket).
I get that none of this is similar to what others are feeling, since if nothing else, experience has taught me that I think and feel in ways that're completely alien to the overwhelming majority of people. But, to relate to others' perspectives, I think back to how I felt when I reached inside to throw one more piece of my soul overboard to stay afloat, and grasped a clenched fist.
Anyways, I'm probably being overly dramatic (and I've had time to reflect and heal anyhow). Point is, I get that peoples' responses are coming from a place of hurt, and I try to give them as much space as possible until they feel comfortable again. It sucks a bit for me, but only insofar as I need to find other spaces to discuss AI topics (as noted elsewhere, the world is not lacking in those) and I need to add another conversation filter to the stack (it's already a mile high, so it's nbd).
I think we missed each other here. It's as @kacey said, what I meant by this is that there's a pressure on people who use LLM's by the "anti-AI crowd" (sorry if this sounds weird) to not use them - or using my original metaphor, to pretend they don't exist.
I'm actually on board with a lot of what you said. I would love to see Scam Altman (not a typo) in prison, along with many others - which, sadly, considering who is in the US government now.... I'm not holding my breath sigh. And I have high hopes for chinese models. With each year they're producing ever smaller-but-capable LLMs that can run in consumer hardware - expensive hardware, granted, but still. My aspiration is to eventually self host my own LLMs and not pay a dime to Anthropic or others.
Thanks. I don't hate the idea of AI, I just hate the companies that are developing it for all the wrong reasons because those reasons shape the way it's being marketed to the public and the use cases they focus on. And I do believe it's important to oppose them not just in spirit but in actions as well.
This is basically where I'm at, with the caveat that I will refuse to use any creative project if it's been made with AI, because that is art theft and I'm not ok with that as an artist.
I think AI has its place in the world, and I'm not against using it to make our lives easier. BUT I am against it being used to justify mass layoffs, doxxing, stealing data, the environmental impact LLMs and data centers have on the world, those who "created it" (used in quotations because we all know that those who are cool with theft of one kind of IP are cool with others) being the be-all, end-all of usage, how it's making people stop using their brains, and basically any other negative side effect of them that I just can't think of right now.
There are appropriate uses of AI/LLMs, and I'm fine with that. Using them to mess with call center scammers? Cool. Using them to help transcribe my doctor's notes and my conversation so something doesn't get missed in my chart? Sure. Do I want AI to have my PHI and use it to make my insurance premiums go up? Hell NO. Do I want creative projects like books or video games, or hell even flyers for my local commuity center events to be made with gen AI? absolutely not.
No I never cared about that. Specifically in the case of code generation I think it's not such a clear cut case. The models are trained on freely distributed copylefted code. Does that mean the resulting AI generated code should be copylefted as well? Maybe. I don't really care if a judge decides AI generated code is all AGPL now. I'd actually be ecstatic if I could make legally enforceable requests to every SaaS company around the world to hand over their now poisoned source code. I'm more than happy to have contributed my public code on GitHub to the model's training data. I've gotten more out than I put in. For image and video generation (which aren't tools I use anyway) there's a much stronger argument against it as so much more of the training data wasn't meant to be freely distributed and copied.
Fair point. And it's a shame because it's mainly an issue of electricity sourcing. It doesn't need to have its current carbon footprint. We should be running 100% renewable electrical grids by now. We are at least making good progress as a species on that. Solar is incredibly cheap, battery tech is improving. I personally wouldn't mind if datacenter expansions were limited by whether they could run off of clean energy.
Depends on the person. I'm able to do new things and learn in the process using these tools. I just finished a project which I will write up and share on Tildes soon. I used LLMs heavily throughout the process (for CAD, programming, data science). I could have done everything by myself, but it would have taken more hours to do so. It's not a kind of project I'd ever done before and I'm quite proud of it.
I'm outside of the software employment world at the moment. Thanks to LLMs I'm hoping to get into a niche area of automated manufacturing soon. I'm highly confident I can build my own CNC machine (not a CNC mill to be clear) given the availability of the basic components/modules/plans online combined with my knowledge and LLMs to fill in the gaps. It helps I have both an industrial warehouse and a makerspace to help me. One aspect I hope to see a tremendous uplift in is patching CNC firmware with LLMs to implement my customizations. Something that could have been a couple weeks of learning will probably be accomplished in a couple of hours. Sure I won't have a personal deep knowledge of the firmware - but I don't think I'll need to. When you build a more standard CNC device you usually don't need to know how the firmware works.
Edit: Honestly if you're someone with technical skills right now I think stretching your capabilities through LLM assistance is a goldmine. Things where you'd typically need to work on a small team (or just be skilled in many areas) can now be accomplished for nearly free as a solo entrepreneur. The LLMs from the top labs are currently subsidized. They only recently got good enough to use in this way. Now's the time to act!
I'm a software dev, and I should preface this with I've been on the rather anti-AI side. I think most of it comes from an inherent distrust of the companies providing it. I haven't ignored it completely. I've tried a few times, and found it something that I have to manage and work around more than a useful addition to my workflow.
I have thought about using it more for personal projects. I have some things I wouldn't mind spending more time on, but am usually too burnt at the end of the day/week to actually do it. But one of my biggest concerns is getting hooked and reliant on these tools to do anything, and then the subsidies finally end. They jack up the costs. Then what? You either pay up, or go back to not being able to use them. I sort of refuse to pay for the privilege of writing code. I don't know if I have a point or a question. Just speculation about capitalism constantly creeping into every facet of my life. It feels awful.
I started to feel comfortable using LLMs more heavily after using the open weight models that are ran profitably. Try GLM 5.3 through openrouter+opencode. Itās a fraction of the API cost of Opus 5 and nearly as capable.
Edit: Consider this subscription if you want to stay away from the shady American LLM companies: https://opencode.ai/go
(seconding this -- IMO Qwen3.8 27b w/a 4-bit quantization even runs nicely locally on a B70 (32 GB Intel GPU). If you have a decent NVidia card with 12 GB+ of RAM, you could probably run one of the older 35b MoE models as well. I have it in the other workspace autonomously troubleshooting a largely undocumented, obscure FOSS project. It's worked well for me so far, at least, but I haven't written anything substantial with it yet)
(I'm not OP)
:fist_bump_emoji: samesies
low key worried that someone cloned me now XD and yeah, been banking on this since late last year, so I'm stoked to see it coming to fruition. Hopefully plans work out and I can meaningfully dig into this over the winter.
(edit) only thing I'd note is that local models have been consistently trailing frontier models by ~6-12 months in coding capability, at least, for a couple of years. So there's a possibility that losing those subsidies -- when the market corrects -- will barely be a speed bump. Fingers crossed ...
Iām planning to start in October. In the end Iāll have a machine that will likely have no equals around the world. Iāll be able to take orders online from a beautiful website with 3D previews and have something produce a real life equivalent. This will take something that normally would involve human labor and do it for cheap. I can lower the price dramatically, expanding the pool of customers. I plan to make multiple websites targeting different product types but all handled with the same machine.
I have actually made some of these widgets by hand and thereās demand. But I realized the prices Iād want to charge are too high. So automation is coming to the rescue.
That sounds really exciting! And yeah, super similar to what I'm aiming for as well (down to the labour price as a factor as well), so I selfishly hope you share some details about your process XD I've always hoped that someone else would land on this train of thought, too, since I've felt quite mad being the only person in the room who's bringing it up.
Well thatās kind of the entire point of automation. I think Iāll also be able to achieve higher levels of consistency, quality, and with less time per widget. I havenāt done anything yet so I canāt say exactly what Iāll do. But you can bet your ass Iāll be using a lot of 80/20 extrusion!
Mmhm! One of a couple leaps of faith from my perspective is (was?) whether the R&D for developing those automated manufacturing systems would be too time intensive/specialized, since that cost of that needs to get amortized into your eventual product. Hence why I'm attempting to move somewhere with a much lower cost of living, in order to buy myself more swings at the bat so to speak š . Additionally, although manufacturing automation/higher throughput machine design is well established at the mid to high scale, it's unclear whether there are unforeseen challenges that would slow down development and deployment at the low end ... there's a whole spiel there that I'm wholly unqualified to talk about, but that I'll hopefully take a bite out of soon enough. Looking forward to 2027! XD
And good luck to you as well! If you're based in America, hopefully you can find a local source for your aluminum extrusion too š
The warehouse I mentioned is actually full of a ton of old 80/20 from a defunct robotics startup. So itāll be free! There are also a ton of cameras and optics there which were made for machine vision.
Sorry if I'm being presumptuous but I think what you might mean here is that the denouncing of LLMs isn't as unanimous as you'd like. I'd say there's a pretty even mix of sentiment about AI on Tildes, with hate edging out non-hate.
The second part of the quote is frustrating. It's blanket assigning a binary sentiment to a large group of people. All anyone can do with that is disagree or ignore it, it's too generalized to be true.
It's far more nuanced. For example, I think AI tools are incredibly useful for some things. Game changing. And I hate that a world changing technology is ultra-capitalized from the start. With the internet we had decades of exploration before the worst of the enshittification set in. That's unlikely to be the case with AI.
And the evironmental concerns are real (if often overexaggerated) and I care about that deeply, I think most people do. I also dislike the potential for AI to contribute to wealth inequality and consolidated control. However, like the internet, AI also increases individual capability. One person, or small groups of people, can accomplish much more than they could before. Sometimes more effectively than big organizations.
Here's the thing: It's not going to go away. It doesn't matter what you think, or I think. If we could somehow burn down all the big AI labs, people would advance the tech in their garages. A new round of funding would launch countless startups in the space. That's because the tech is genuinely useful. There was a lot of emotionally motivated writing about how LLMs don't really help anyone, and anyone who thinks they do is just deluding themselves, in 2024 and even early 2025 but that's painfully outdated. It's only people with no expertise in any of the fields where AI has value that are saying that kind of thing now.
I have nothing against a website listing AI free options. I think it's great, bring on the demand for artisanal, entirely hand crafted, options.
But also the reality is that, in the field of software engineering at large, AI assisted (as opposed to wholly vibecoded) development is already the rule rather than the exception and nothing is going to change that.
You're right about this and I apologize. I've been too angry about the topic lately, for a number of reasons, and I let that emotion take the head of my discourse. It's not in the spirit of Tildes and could have been seen as uncivil, so while I don't agree with the responses, I appreciate that they were more level-headed than mine.
I do stand by the fact that I can't support AI-assisted development (or genAI in general), regardless of how commonplace it's become. I do think that, in a different form, I would at least be able to accept it. My problem is with OpenAI, Anthropic and others relying on or producing and maintaining behemoth models in ways that have no business being tolerated, but I've had to remember we were already studying what we know as AI before 2022 and there's people using the tech to do fascinating and useful things without relying on scraped data and misuse of energy (even for art! I learned how SynthetizerV and similar products do their work recently and it's a great example). The concept itself isn't immoral to me.
There might be a day where it's possible and viable to get assistance from a local SLM trained only on data that was fairly obtained from consenting authors (I think it's even likely). That, I likely still wouldn't use unless necessary, because I've been down the road of using LLMs for programming and found it hurt my competency more than it helped; but I would at least be okay with its existence.
Thanks for the civil response! No argument with you choosing to avoid LLMs. Even in software engineering I suspect there will be room for competent people who don't want to use agents for... well I don't know how long, but maybe quite a long time.
I hope it gets less frustrating.
A lot of these issues feel like they're because we're in the transition from the "legacy" world into the post-AI world. During the transition, everything feels broken. There's an adaptation. It's pretty obvious with the current pass that's happening in cyber security.
In terms of stolen work, since I've started learning about Linux and free software, I've always felt like the current system was broken. https://web.mit.edu/m-i-t/forums/copyright/index_transcript.html The culture change that AIs bring seem to me to be something that will be positive.
For the environment impact, I guess I got magical thinking, but I feel like it's the only way to speed up research to save the earth.
In terms of benefit, I'd disagree, the company I work for, for example, uses it to save lives.
I don't think my response is that good, but I didn't have time to think it through fully. Will probably revisit it later. These are my general ideas.
LLMs as they are right now (opaque closed-source models, tightly controlled and guarded by billionaire-owned corporations) are the complete antithesis to FOSS ideals. They are an extreme shift in the opposite direction of what you're implying here.
You are correct, this is 1000% magical thinking. Even if it were feasible that predictive text generating engines could somehow come up with a solution to our climate problems, none of these companies are currently or will ever bother actually putting any kind of meaningful resources behind that kind of research, because it won't increase their profits (and I mean, realistically they probably know that the idea is absurd to begin with).
The thing about the frontier models is that theres no real secret sauce. They advance mostly based on throwing raw compute power into training, which is why all of the frontier labs have similar performance. Those models get distilled, and cheaper Chinese and open source models see similar performance within six months to a year. Openai, anthropic and Google are controlled by billionaire run corporations, LLMs as a whole aren't though.
One of two things is going to happen. We're going to see increasingly diminishing returns, where in order to get any improvement, more capital and power than exists on the planet is required, and open source competitors will close the gap and catch up. Or, these models will never stop getting significantly better, in which case we're likely fucked as the technology dominates the human race.
I think the former is the more realistic option, but I don't see anything besides harsh multilateral government enforcement of a total ban being feasible if it's the latter.
The current versions of LLMs are opaque, but there's a consensus that we're in an AI bubble which could pop at any time. The open models will be run on cheaper and cheaper hardware overtime. But I didn't mean it that way only, there's a part about how it's a good tool for compacting knowledge. Also last time I checked, the output from LLMs couldn't be copyrighted.
It's not the predictive text that would necessarily discover something of value (though in math they are making some pretty significant breakthroughs without much guidance), but its ability to speed up research and how researchers use it as a tool.
If I were you I'd be concerned it's a way to subconsciously bury my head in the sand, as there isn't any research required (which, to be clear, isn't the same as saying we shouldn't do any). Perhaps the biggest impediment to environmental progress is the lack of any consensus that we should make progress. In this I'd say AI is on balance an impediment, as well as being environmentally detrimental directly.
It is the current system that affords authors the right to license their work under whatever terms they please. There is no Free software, with its strict licensing requirements, without a legal system that protects these rights.
You'll be less pleased to know that one frequent use of LLMs is to perform so-called copyright laundering of Free software. Modules, libraries or whole applications are fed into the LLM which rewrites them ever so slightly so as to rid the software of signs of the original authorship and, at least ostensibly, the licensing requirements imposed by the original author. The idea is that by doing this you can more easily create proprietary software based on Free software.
The negative effects of LLM use on the maintenance of Free and Open Source software are well known. For every useful contribution to a popular project there are a hundred patches and PRs by useless charlatans who practically function as transparent proxies for the LLMs they use by uncritically reproducing their hallucinations, neither understanding the problem they may or may not be solving which may or may not exist nor understanding the solution they submit. This has made many F/OSS project maintainers more restrictive in what kinds of contributions they'll accept or entertain at all, potentially throwing some babies out with that dirty, smelly ocean of bathwater.
We know what the problem is and we know what we are doing to cause that problem. The reluctance to solve the problem accordingly is a political phenomenon.
Except that free software only exists to counter the current system. There wouldn't be a need for free software licenses if we didn't have to fight fire with fire. You can read more about it in https://www.gnu.org/philosophy/free-software-even-more-important.html.
I agree that LLMs can be used to launder free software. My idea above was for use in the opposite direction. You can use an LLM to recreate currently closed source programs. (Last week I got my own implementation of a software that's closed and requires admin rights and runs permanently as a background service. Now that I have the code, I'm testing to make sure that it works well before I release the code.) So I'd say you can also create more software in general and it works both ways.
While the negative effects are well known in open source projects, I see them as being part of the transition from our legacy world. It's allowing more people to contribute even though they don't have the knowledge to. That has negatives, but we'll learn to deal with it and people will get better. I think Linus Torvalds' stance on this is the right one.
Knowing the problems on environnement and being able to fix them is completely different. The only thing that seemed to work so far was during the pandemic lockdowns. Most people don't want to go through that again. We're already locked into bad outcomes and predictions are happening faster than expected. Vicious cycles are going to get worse and worse. I don't expect anything to get better the way things are going.
It's not so much a matter of opinion on what LLMs can and can't do as an observation of an ongoing phenomenon.
Yes, plagiarism is to be expected everywhere, not just of Free software.
What does this even mean? What is a "legacy world" and what are we transitioning into? Will the world you think we're transitioning into not have a legacy? I try to think clearly about this issue, but terminology like this isn't making things easier. It seems to me like a creative alternative to waving your hands instead of presenting an argument.
With that perspective in mind, the hope in LLMs magically solving the problem for us makes even less sense.
Legacy to me is how there are many issues with the world right now. A lot of those issues are invisible. You can look at cybersecurity, the ways AIs can get into systems already existed. The bugs didn't appear because of AIs, they only get exposed by them. The speed that they get exposed increases as AIs improve. I say legacy because at some point, there should be fewer and fewer issues to patch. When that happens, that's when the transition is done. While it's happening, it looks like it's the end of the world for cybersecurity. But once you're on the other side, the world should be better since there are fewer bugs.
The same thing can happen with everything else. When you can't trust the news online, perhaps we go back to local news that can be verified easier.
It think it's exposing issues that were already there and that we have to deal with.
On plagiarism, I see it more as unlocking knowledge to the whole world which is a good thing. Calling it plagiarism implies that only one project of any kind is allowed to exist at once.
That "when" assumes a particular answer to a very big "if". I'm seeing no indication yet that LLMs will ever be used to solve bugs faster than they are used to produce them. The economic incentives are pointing elsewhere: access to models that can thoroughly audit and produce high quality threat analyses that identify critical vulnerabilities is much more expensive than access to models that can produce poor to mediocre code which ostensibly implements solutions to business domain problems but has an error rate similar to that of a junior developer.
There are a few differences between the LLM and the junior developer that further exacerbate the problem. The junior (without an LLM) is limited in their productive capacity by their lack of skill, knowledge and wisdom. They will produce code at a lower rate than the experienced developer, because they will have to stop to learn things more often, ask more questions or will simply avoid problems that seem challenging. The burden imposed on more skilled team members in reviewing and auditing the low-quality code is consequently limited. The LLM knows no such limits and will spit out massive changes without ever gaining knowledge. I think people working in engineering organizations that are being pushy about LLMs can corroborate my experience that this eventually results in lower quality reviews.
I worked in an organization where management "addressed" the problem by reminding developers that their code is their responsibility, whether they wrote or generated. That sounds nice in theory, but is ultimately simply not true. The resulting problems won't magically be directed at the person who created them, but at the team that maintains the software.
You should consider that what works for Linus Torvalds might not work for every other software project. Linux has generally high standards of inclusion because it deals with the production and maintenance of software responsibly and uncompromisingly as a high-stakes engineering problem. Being as widely deployed as Linux is also attracts a lot of attention from security researchers and skilled developers, funded by organizations with legitimate interests in its quality. I am however using (and I'm being used in) a lot more software produced by organizations with much lower standards, driven by the market to "get shit done", "move fast and break things" or whatever other slogans they use to justify being quick to market at the cost of quality of engineering. This use of these systems is often not voluntary.
Meanwhile, with the advent of LLMs, the difference in cost between producing such low quality systems and producing high quality systems has increased. I think it will increase further as investors become impatient waiting for the chatbot vendors to realize the potential they see in them as profitable businesses. If any charlatan with no knowledge of engineering can use an LLM to quickly generate volumes' worth of low-quality software alone and something with the rigor of Linux still demands the attention of hundreds of engineers on the LKML, the production of software will be much more in the hands of incompetent charlatans than people like the Linux maintainers.
Itās a bit ironic to complain about ācopyright launderingā when free software got it start by reimplementing proprietary software. Stallman wrote copylefted code that reimplemented proprietary Unix commands that did the same thing. This was done specifically to evade the licensing requirements of the original software!
And to be clear, I think implementing free alternatives to proprietary software is a good thing, though the software vendor might not like how it undermines their business model. If youāre trying to make a living writing software, competition from free alternatives is a problem. (Free only after theyāre implemented of course. Actually building the software still costs time and money, though less than before.)
Sometimes people hoped that free software would become āviralā due to network effects along with copyleft restrictions. But to the extent that it has, itās mostly not due to the copyright restrictions. Every popular programming language has a huge collection of open source libraries available and copyleft doesnāt play much of a role in that. And now that AI-driven rewrites are feasible, itās even more dead. Itās too easy to make clones that are free of whatever copyright restrictions you might add.
The remaining āmoatā is basically trust, along with inertia. Just because some unknown author published an AI rewrite and it seems to work, would you trust it? How exactly do new projects become trusted? We often depend on network effects - if an open source library has been around for a while and seems to be used a lot, maybe itās okay? We donāt even know the names of most of the developers whose work we depend on.
When a new version is published under the same name, hopefully upgrading is safe? Nowadays package managers are adding ācooldownā periods so that ordinary developers donāt upgrade immediately. There are organizations that automatically scan new releases for malware, and adding a delay gives them time to sound the alarm. I wonder to what extent these vendors are starting to use AI for their scans?
That's genuinely terrifying to me.
As a wee note, afaik the only research area of note is direct air carbon capture. Everything else seems addressable with current day technology, but it runs into social problems (people despise change) and class problems (currently wealthy folks operating companies dislike the prospect of uncertainty or reduced profits). R&D expenditure seems minimal beyond stuff like advancing fusion energy research, but that is slow because it takes a long time to build reactors and validate theory afaik.
Thereās also various forms of geoengineering (mostly in the area of solar radiation management), but those are a lot more controversial. As far as direct air capture, thereās not really technical limitations. Thereās already pilot plants running. The real issue is cost/efficiency (they are massive energy hogs), which might become moot with either cheaper overbuilt renewables or micro-reactors (especially if they can cogen and provide process heat).
Afaik, the primary concern is that DAC plants constantly run into fouling problems with their catalysts. An AI could -- hypothetically -- scan through potential alternative catalysts which might be more durable, or otherwise automate the testing process at small scales (similar to how we do microfluidic assays) in order to rapidly assess viability for groundbreaking approaches. Since right now the problem with direct air capture is that it'd need, like, some absurd double digit percentage of the world's power generation to make a big enough dent to be viable (as you note, but hence the need for advancement). In other words, overbuilt would have to look like blanketing the Sahara in solar panels or a breakthrough in fusion reactors.
Er, all of which is to say, what're currently referred to as DAC technologies are unlikely to advance much, afaik (circa like ~2022 which is the last time I did any reading on this). (edit) (for context, from my non-expert, limited research perspective, wood vaulting or some twist on ocean fertilization are the most promising technologies we have for carbon sequestration)
Furthermore, w.r.t. solar radiation management, I think we already know as much as we can without experimentation around stratospheric sulphur aerosol injection which afaik is the only practical method anyone's proposed. Fingers crossed that space mirrors work out, but I'm not banking on it ...
You're getting a storm of pro-AI responses from the looks of things, so I just wanted to let you know I really appreciate your post. I've long felt that any time I enter a conversation with someone pro-AI here that it feels like I'm talking to a brick wall that is determined never to accept any of the arguments against AI usage, and it drives me nuts.
From my perspective, it feels like an automatic defensive measure; people got interested in the tech for one reason or another, and now that they've gotten used to it they don't want to switch back. So they find the reasons they can to justify it against the people telling them to stop, however poor they may be. If this is an accurate assessment, I can still have hope that models will eventually force them to re-evaluate when prices inevitably jump, but damn if it isn't frustrating.
Just want to provide the info that there are models served at a profit right now from Chinese vendors and theyāre more than good enough for professionals to use for building software.
Interesting! I have less reason to be optimistic about the future, then.
Trust meā¢ļø?
Anyway, not for me. I don't have an issue with AI assisted coding.
Vibe coding though, yes. Different story. I avoid those projects by default.
Whatās your personal distinction between the two?
I'm not who you asked but here's mine:
AI assisted:
I designed the code, with or without iterating in a solution with an LLM or another human. I may have accepted "autocompletions" provided by an LLM, or used a tool like Claude Code to describe a function and accepted a whatever it generated (after reviewing it and understanding how it works). I may even so large scale refactors using the same tools, but all under my review and guidance.
Vibe coded:
I described an end goal and let an LLM agentically / iteratively design and write it autonomously with minimal review or understanding of the underlying code.
The former is using tools to speed up my own workflow.
The latter is outsourcing entirely and blindly trusting something works as intended and is of good quality.
The way I see it is: it's the exact same as Google or StackOverflow. We don't say software is "Google assisted" because you searched online to find an answer. LLMs are the same, but much more specific and faster. It can parse out documentation in minutes that would normally take hours.
Iāll add another thought: my coworkers with software development backgrounds using AI to help them would be AI-assisted coding. Me (with a very poor-understanding of coding and programming) using AI to spit out a program would be vibe coding.
Purity test, if you ask me[1]. Yes, of course, Vibe coding is bad, nobody's contesting that, but there's nothing wrong with using AI in your coding projects provided you could conceivably do the same work yourself, given enough time. To me, Linus Torvalds was correct on this issue.
I really feel like these project lists, but especially Codeberg as a whole, wants to desperately feel better than everyone else. And you know what? Fine, they're better. Gold star. You get to go to heaven because you didn't use it. There you go. Congrats. But provided the project is good and the operator knows what they're doing, there's no substantive difference between good hand-written code and AI-assisted code.
[1]: Also evidenced by the fact that to get on this list, you need to post on Mastodon about your project. What, no Mastodon? No ActivityPub? Well, then you can't be serious about open source software, clearly.
For those interested, here's a useful thread on various harms.
I understand disagreeing with this, I don't understand considering it an attack on your person in any sense.
I think a "thought terminating cliche" while not the term I would use, is very different from "an attack on my person" and anyone who agrees with you.
I think accusing someone of attacking you is serious enough to make this point despite it being tangential to the actual conversation. It certainly doesn't bring the temperature of any discussion down. And if your cynicism is impacting your experience so much as to accuse a link to be a personal attack, then I would suggest disengaging, even if only temporarily.
I wonder if they will also refuse to use AI to scan for security bugs?
Not to worry, someone else will do it for them.
You talk about this like it's some infallible error-finder, and not something prone to contradicting itself each time it's run. You also talk about it as though doing this has no harms to weigh against.
Why would it need to be infallible? Code review isnāt infallible either, but itās still a good idea.
I'm not getting into a debate with you on the utility of AI. I'm pretty sure we've done that before and it became abundantly clear you were not going to change your stance, and I don't think I've ever seen you do anything but defend AI besides.
I'm just pointing out why the confidence and snark in your initial comment was undue.
Well, finding real vulnerabilities with some double digit probability for pennies, that others now can also discover for the same price doesn't sound like a bad deal. Certainly cheaper than getting all your data leaked/hacked.
I let myself get tricked into responding to the top comment and then deleted it. Good little website. Happy to see GIMP on there.
An index for software that is clearly marked that is made without the use of generative AI.
I wonder how these projects would stand up to the watermarks that are being applied. Itās one thing to say, and possibly believe, that your project doesnāt have any LLM code in it, but if there are unknown outside contributorsā¦ and of course the library ecosystem is riddled with LLM code, so if you vendor any of those libraries you may unwittingly add it in yourself. Itās like microplastics these days.
Cute but the whole Git platform is vibe coded at this point I donāt know how anyone would actually avoid all vibe coded platforms.
Also, the software to tell teachers if a paper is LLM generated kinda sucks, I donāt see how this could be better.
Ive been collecting used books because anything published after about 2023 was at least edited by an LLM, if not written by one, and theres no way to tell for sure other than just trusting everyone involved.
What Git platform? Git is just a version control tool. Did you mean Github?
All of them, github, bitbucket, gitlab, every single service that hosts git repos is vibe coded now.
So if youāre trying to use Git, youād have to set up your own remote if you wanted to avoid vibe coded projects.
On top of that:
Microsoft services are vibe coded, Google services, Apple services, your banking apps and websites, Doordash, Uber, Lyft, all online ordering apps, all social media, all entertainment services, all that shit has been vibe coded for the past two years.
I know the point is probably so that when youāre building a project you can avoid using other peoples vibe coded libraries because they may be more prone to breaking, but if youāre gonna do that you gotta host your own Git remote and I donāt even know what flavor of Linux has no vibe coded pieces in it. Everything has this shit in it.
Not Codeberg/Forgejo, which is what this site recommends.
Minimum remote for git is a directory, possibly accessible via ssh.
ie: git+ssh://192.168.1.10/path/to/repo.git
A remote can even be a directory on the same machine. I did both of these recently to easily move code between my IDE projects dirs to something's plugins dir for a couple of things I was working on.
Git doesn't require any third party code hosting services to operate or exchange source code.
These services provide a (hopefully reliable) storage location and a web ui for the managing access & review, among other things.
You wouldn't want your Git remote on the same machine, though, and any cloud software is gonna be vibe coded.
Both git and Linux allow AI generated/assisted PRs.