I'll be honest, I'm just kind of sick of these takes. Don't misunderstand, please - I also think AI is making the world worse in the long run. I do however think that the layman - and I'm quite...
Exemplary
I'll be honest, I'm just kind of sick of these takes. Don't misunderstand, please - I also think AI is making the world worse in the long run. I do however think that the layman - and I'm quite comfortable calling Eddy Burback a layman on this subject - is simplifying and confusing a few things.
First, though, a quick glossary - I'll call marketed solutions like Claude, ChatGPT, Gemini, Midjourney and DALL-E "AI", and the underlying technology "machine learning" or "ML", so we can have this conversation without mixing up what torment nexus the capitalists are parading around with the actual research of actual scientists. And to give you a TLDR: AI bad, ML neutral but interesting.
I don't think machine learning is a bad technology. There's certainly an argument to be had about what ethical data collection for training purposes looks like, but I am neither knowledgable enough on the topic of copyright nor do I personally care to litigate this issue - but saying that there's "no application for any of this" and just calling the models and technology "bad" because the output isn't very high fidelity is just lazy. I've been following (and tinkering with) diffusion and transformer models since the mid 2010s. I personally was quite impressed when ChatGPT was released. GPT 3.5 and its contemporaries were incredibly competent compared to other solutions available at the time, and I was hopeful to see what could be done with them. Sure enough, a revolution was sparked. Suddenly, everyone worked on a cool ML project. Auto-GPT comes to mind, and the stuff that Stability AI was doing on the front of open-source image diffusion was really interesting to watch and keep up with. Seemed like there was a new breakthrough every other week.
The capitalists of the world however didn't look at a Stable Diffusion or an Auto-GPT and thought "whoa that's really neat, this teaches us so much about how computers and maybe even our own brain works, this is obviously not ready for prime time nor will it be in the forseeable future but just making computers do this at all is insane". They looked at a mediocre copywriting tool and thought that they can sell this.
And sell they did. I don't have to look far to find corporations including "AI" in products where it doesn't even make a lick of sense - that AI grill from CES comes to mind, and Google even replaced Assistant with a new AI powered version that struggles to set your pizza timer, but does recommend you put glue on it. This is failure, obviously, and the execs didn't understand the technology enough to consider this was a bad idea.
I'm not saying that Gemini is bad because it tells you to glue your cheese on. I'm saying that Gemini and its contemporaries are actually very good at the things they are intended for. This includes sentiment analysis, summarisation, that sort of thing. An example here: Kagi's Universal Summariser is an excellent tool, but it's not good at deterministic tasks like mathematics or counting the number of Rs in the word "Strawberry". And yet, that's what we see when people dunk on ML. "Look at this stupid idiot, can't even count right" you will hear in the echochambers of especially artists - more on that later - as if that was ever the point of the invention.
I think AI is a mistake. I don't think ML is a mistake. I think the technology should have been left to us nerds, people who know what they're doing and know the limitations and quirks of this (in this iteration) very new and unexplored frontier in computer science. And to that end, I don't think Eddy is even that wrong in the video - to quote Theo Katzman, I don't really want to sing along if a computer wrote the fucking song.
But who would even make a computer write a song with the intention of it becoming a hit? Hint: It's the guys with the dollar signs in their eyes. The technology itself is in my opinion completely neutral. Machine learning will not take your jobs, and machine learning does not steal from artists. People do. The people that choose to abuse these research efforts for their own personal gain, laziness or convenience. That's the problem. Not the computers calculating matrices.
I'm not sure I understand your point. On one hand, you agree that the current iteration of AI as marketed towards end consumers sucks. So what issue do you take exactly with these takes? That they...
I'm not sure I understand your point. On one hand, you agree that the current iteration of AI as marketed towards end consumers sucks. So what issue do you take exactly with these takes? That they are all too similar?
Don't get me wrong, I don't disagree with you at all. I think ML has a ton of credible uses, and I also agree that right now, beyond tech demos, it doesn't really offer any valid use to the end consumer. I think it can be used in varying ways to enhance products that are used by end consumers, but I really don't understand what AI offers to you in its current state.
There is one exception; character art for TTRPG characters. Most of these are small groups of 3-6 people and commissioning art of your character is something that you do only when money is no issue or you have spent a particularly long time with a character. Paying for good art is expensive, as it should be, you're paying for someone else's creativity.
While the way that current AI image generators are trained is media piracy, I don't consider this specific usecase too problematic, because the overwhelming majority of people generating art fo their TTRPG characters never pay an artist to make them "real" anyway.
What I find entertaining about the current situation and the hype behind AI is less about the technology itself, and how an army of business and marketing MBAs are fevereshly working to sell you a product that is obviously flawed and has a miniscule amount of any real usecases, and how the best they come up with is an app that improves business communications when the issue I have with a lot of these communications is that this email didn't need to be better worded, this email didn't need to happen at all.
Or the Gemini ad about a father having the AI compose a letter for his daughter to send to an athlete. Like that's what you come up with? Who made this? One of the most endearing things a parent can do; sitting down to compose a letter with their child for their greatest idol, and your idea is to abstract it into a fucking chat prompt? Who made this? What room of, likely creatives working in the advertisement industry, thought this sounds like a good idea? Was there even one parent in the room?
AI advertisement presents everyone in it as a clueless idiot who can't decide on anything. It is, at it's core, the ad industry doing it's best to sell you a shit product. Paradoxically, AI is also a huge bubble right now, and more people are throwing more money at it than anything else, because modern free market capitalism needs it's quarterly growth or it's fucked, apparently.
Yes, maybe I wasn’t clear. Sorry about that. My issue with these takes (and let’s be honest, Eddy isn’t anywhere near the worst offender) of “AI output bad therefore AI bad” is that it’s...
Yes, maybe I wasn’t clear. Sorry about that. My issue with these takes (and let’s be honest, Eddy isn’t anywhere near the worst offender) of “AI output bad therefore AI bad” is that it’s narrow-minded and does nothing to address the real problems with AI. The hordes of none-to-underpaid Indian workers on Mturk classifying images. The moderators that have to sift through all the heinous shit online. But no, AI is bad because it’s “annoying” and “steals from artists”. It almost pisses me off. Like, the Waymo segment. Do you realise how incredible it is that this tech works at all? But no, let’s reject it because it makes known exploiters Uber and Lift obsolete and because my Waymo didn’t understand navigating a parking lot. There’s good and valid criticism of both AI and ML, but this ain’t it.
I would like to stake my claim that it does steal from artists. Training an ai to create works of art based on others peoples works of art without their consent or fair recompense is stealing. You...
I would like to stake my claim that it does steal from artists. Training an ai to create works of art based on others peoples works of art without their consent or fair recompense is stealing. You know that if the situation was reversed, these companies pulling the bs wouldnt think twice to litigate in response. If i took all of microsofts ip and ran it through ai image gen to make and sell my own products, im sure Microsoft would call the lawyers when they found out. But i digress because that is up to interpretation.
What i find not to be up to interpretation is that it is stealing jobs from artists already. Big companies are already using it to do art for promotional material. See this debacle with microsoft. Thats an artist that lost a gig, or even an employed position at another company. Here is a video about a professional graphic design artist getting layed off after find out his company had been training an image gen AI on HIS art and work for months or years. He was literally replaced by an AI that is designed to copy his style. Those are just the first two examples off the top of my head, but this going to get worse. Publicly traded companies will take the path of least resistance, and paying a licensing fee is far cheaper than employing a real human being.
I also find your defense of waymo to be a bit dismissive of the workers it would replace. Known exploiters they may be, Uber and Lyft did create jobs for a lot of people that were otherwise struggling to get anything. It may not be a good job, or a fairly structured job, but its better than nothing. Real people still see real benefit from being employed (contract workers, not proper employment) through them. Nobody sees any benefit from waymo except the people at the head of the company. You are taking these jobs away from already vulnerable and exploited people. I am electronics-pilled guy, and no matter how many times i get impressed that waymo cars are real and functional, its always swallowed by the fact that in the end, the only meaningful thing these cars change is putting another human out of work.
Whenever I come across this line of reasoning it makes me think of how I've read that people were very skeptical of automobiles in general around the end of the 19th century, due at least in part...
no matter how many times i get impressed that waymo cars are real and functional, its always swallowed by the fact that in the end, the only meaningful thing these cars change is putting another human out of work.
Whenever I come across this line of reasoning it makes me think of how I've read that people were very skeptical of automobiles in general around the end of the 19th century, due at least in part to how it would put horse-drawn carriage drivers and repairmen out of work.
I'm of the opinion that jobs becoming obsolete due to the advancement of technology should not be used as a reason to rally against or artificially impede that advancement. Instead we should be thinking about how to provide other opportunities to or otherwise help those people who are out of work, ideally moving towards an eventual post-scarcity society where nobody needs to work to live, and instead everyone can use their time to pursue the things that they actually want to be doing. We'll never get there if we prioritize outdated human-powered technologies over better automated alternatives just so people can stay locked into jobs that they would (in many cases I'd guess) rather not be doing anyway if they didn't have to.
I'm sorry to say that I just disagree, even if I'm pedantic about it. Taking copies of someone's work to train a model on is not theft, it's copyright infringement. Theft is depriving someone of a...
I'm sorry to say that I just disagree, even if I'm pedantic about it. Taking copies of someone's work to train a model on is not theft, it's copyright infringement. Theft is depriving someone of a physical good, and the originals still exist in this case. The reason why I care to differentiate is simple: Obviously I get what you're saying, you're saying that it's as bad as theft and that's fine, we can probably agree on that, but saying it is theft then creates a disconnect - if it's so bad, why do you have to pretend it's something else? Dr. Stuart Ashen said this quite well: You can think Athlete's Foot is as bad as having an ear infection, but if you say that Athlete's Foot is an ear infection you're just going to confuse people. Same with this. If you want to litigate whether or not this practice should be illegal, you need to make your case within the current vernacular of the truth, for lack of a better term.
As for your other two points, I don't think you understood what I wanted to say. I'm sorry that happened to Nadestraight, but that just sort of proves my point. It was the corporation that betrayed him and copied his work without permission, not the AI model. Humans made that decision, not a machine. This is, to the company, functionally no different than outsourcing an IT department to India because it's a fraction of the cost as the in-house guys that are then laid off. Nothing new, and not the fault of a new technology (I'm going to go out on a limb and say that you probably wouldn't campaign for oursourced Indian labour to be illegal) - Again, AI vs ML in my original post. Same with Waymo. Originally, it was a research project, and now it's a product, because humans decided it can be sold. I don't think I can fault the technology for that - and for what it's worth, I see this more as a "changing of the guard" than a death knell for Uber and Lift. The taxi lobby kicked and screamed through that transition 10 years ago as well, and yet it happened. That's to be expected in capitalism, so if we agree that that torment nexus is here to stay, we also have to shrug and accept this change.
This current implementation sure. But there is the pipedream that we'd have a proper city wide/country wide/continent wide network of cars which could provide as good if not better alternative to...
no matter how many times i get impressed that waymo cars are real and functional, its always swallowed by the fact that in the end, the only meaningful thing these cars change is putting another human out of work.
This current implementation sure. But there is the pipedream that we'd have a proper city wide/country wide/continent wide network of cars which could provide as good if not better alternative to private car ownership.
Especially in sparsely populated areas where it doesn't make sense to have a local Uber/taxi network this would provide an equalizer compared to big cities.
Of course I wouldn't trust Google to do that but there's potential beyond just replacing drivers.
We already have an as good if not better alternative to car ownership, and have for hundreds of years. Public transit. A fleet of robo cars that shuttle people to and from where they want to go...
which could provide as good if not better alternative to private car ownership.
We already have an as good if not better alternative to car ownership, and have for hundreds of years. Public transit.
A fleet of robo cars that shuttle people to and from where they want to go sounds great, because you're imagining a fictional best case scenario where everything works perfectly. In real life, these things will be rolled out as soon as the massive corporations creating them decide they're good enough to be profitable, which means they'll be as expensive, dangerous, inconvenient, and annoying as the market can bear. As evidenced by online advertising, social media, crypto currency, ride sharing, and every other "disruption" innovation of the past 20 years, that's quite a lot.
Local and state governments could, right now, change policies to incentivize transit and pedestrian oriented development, stop subsidizing private car ownership as much as they do, and start subsidizing walkability and public transit, and within a very short time, people's lives would be improved via tried and tested technology that's existed for hundreds of years while tackling the biggest contribution to climate change. We don't do that, not because the technology isn't there, but because private companies can't get as rich off of it.
That's the most frustrating thing about transportation when people talk about it as if it's not a problem that's already been solved in many places.
Unfortunately, with the incentives currently in place, robo cars will only ever take off if huge companies discover they can make an absolutely ridiculous amount of money off of it, which pretty much precludes it from ever being a good deal for the average person. Even if alphabet's intentions with waymo are as pure as can be, the auto industry is such an entrenched and valued part of the American economy that they will spend every last lobbying dollar they have to ensure they get their cut in order to "defend American jobs" or whatever the justification is, and we're going to be right back at square one.
The problem of private car ownership has never been technological, it's always been political.
Yep, I use public transit too and in addition own a car. It makes everything way more convenient as I don't live in the city center so everything takes time with the switches etc. I'm not going to...
Yep, I use public transit too and in addition own a car. It makes everything way more convenient as I don't live in the city center so everything takes time with the switches etc. I'm not going to explain more than that about my personal life because it's nobody's business but I'm pretty sure lots of people can relate to using and having both options.
And I'm aware that big companies are for profit that's why I mentioned that I probably wouldn't trust Google on this.
But a man can dream of a world where robotaxis would be implemented as a complimentary service for public transit.
Where do you draw the line between end consumers and enhancing products used by end consumers? I pay for ChatGPT and apart from various programming tasks I've used it to write a semi-scientific...
right now, beyond tech demos, it doesn't really offer any valid use to the end consumer.
Where do you draw the line between end consumers and enhancing products used by end consumers?
I pay for ChatGPT and apart from various programming tasks I've used it to write a semi-scientific paper, where it transformed my notes part by part into written chapters that needed a lot of editing, but it cut my time by about 50% (I was still paid for 100% of the time). This seems like a specialized end user application.
The first professional early adopters of Midjourney were concept artists who often drew over its output because their creations don't get published as is anyway. I recently used Dall-E 3 to create a cool vintage logo to be used as an intro for my band's live video recording. I had to work on it for about an hour to fix spelling and typography errors because Dall-E is a bit outdated now, but it's creative and I couldn't get the same level of creativity out of Flux, which does text perfectly. I suspect Midjourney would do both, but I didn't want to pay for it - ideally the process would be just to take it and use it as it is, since it's only shown for like 10 seconds anyway, and I think we may be at that point already. That seems like end consumer usage.
Last week I talked to a guy who works for a company that offers finetuned translation LLMs adapted for specific subjects. Imo that is one area that is clearly not a bubble (and that is going to have bad consequences for an already underpaid field).
AI assisted search engines I guess are a typical example of enhancing an existing product, but some of the results are so novel that they pretty much are new products, like LLMs searching through released scientific papers and giving reasonably accurate summaries with citations to specific papers, because this lowers the barrier to learn something in a subject where you lack formal education enough that it allows more people to do it. I use this and it has been very practical for me.
And the "non-AI" applications of machine learning are even broader: for the live music recording mentioned above we used a tool that splits a stereo recording into individual channels, so that you can do a mix of something that wasn't recorded in multichannel. It now works well enough that it's actually useful, although imperfect, and it singlehandedly saved that recording for us. I used an open-source tool and most online services are just an application of this same open-source tool (but paid, that is undoubtedly a bubble), but some are not and they will probably continue making money because it's simply useful for musicians not just for recording but also for learning new songs, for people who do remixes and mashups or for karaoke.
Rereading the paragraph you're quoting, I didn't phrase myself clearly. In the same vein as the person I was responding to, I divided AI (as in, chat gpt and others) and ML into two broad...
Rereading the paragraph you're quoting, I didn't phrase myself clearly. In the same vein as the person I was responding to, I divided AI (as in, chat gpt and others) and ML into two broad categories to keep it simple. I agree that ML has quite a few uses. I friend of mine uses the audio splitter for songs to create mashups a lot, and I helped them set it up because they're not too great with computers and running python through cmd was difficult for them.
I've not had luck with AI assisted search engines so far. The famous pizza glue incident or that one time Google's Gemini recommended suicide come to mind.
The core of Burback's criticism remains, though. Where is all this efficiency going? Not into our pockets, that's for sure. You think if those concept artists become more efficient, they're suddenly going to get paid better? As freelancers, maybe. Worst case it's going to lead to an expectation from employers and soon enough you'll need to use it to keep up with demands.
I'm happy that chat gpt works for you, I personally cannot use it at work because legally it's unclear who code would belong to. I've never found the writing too compelling because there is no purpose behind it, and I like writing more than editing. Even if I'm more efficient, I'd rather write it myself. And that's what I mean: I'm not interested in watching over a robot doing stuff. I would rather do stuff.
This is mildly off-topic, but I recently started using perplexity.ai for searching through papers. It sometimes gets the summary wrong, but it's better than google or bing and it almost always...
This is mildly off-topic, but
I've not had luck with AI assisted search engines so far. The famous pizza glue incident or that one time Google's Gemini recommended suicide come to mind.
I recently started using perplexity.ai for searching through papers. It sometimes gets the summary wrong, but it's better than google or bing and it almost always gives me a solid place to start (I always have to check those papers anyway). I use the free version, so feel free to try it if you ever need something like that. Pizza glue etc. imo all come from companies releasing significantly rushed products when trying to outcompete OpenAI.
I friend of mine uses the audio splitter for songs to create mashups a lot, and I helped them set it up because they're not too great with computers and running python through cmd was difficult for them.
Recommending Ultimate Vocal Remover, it's a free GUI for common models, it's pretty good, doesn't have any external dependencies and downloads the models through its own download tool.
I've never found the writing too compelling because there is no purpose behind it, and I like writing more than editing. Even if I'm more efficient, I'd rather write it myself.
It depends, for me writing an article was just a job, and one of the parts I dislike the most is the beginning of an article or beginning of a chapter, the introduction and planning how to structure the information from my notes that I want to include there. ChatGPT completely removed this obstacle and even if I had to rewrite 75% of what it gave me (most of the time it was less), it was faster and more pleasant for me to finish it. Not discounting that this is a subjective thing, of course.
There's a bizzare trend of advertisements that insult the consumer going back for practically forever. There are yogurt commercials that pretend that men are idiots. There have been coffee...
There's a bizzare trend of advertisements that insult the consumer going back for practically forever. There are yogurt commercials that pretend that men are idiots. There have been coffee commercials that say that a man will leave a wife if she doesn't make coffee good enough. It's basically like pickup artist negging. As much as people want positive things, the thing that really makes people motivated to do things is the possiblity of losing something. So as a result many ads try to capitalize on FOMO.
I think I hold 2 salient points that back up a lot of your thoughts: ML is pretty much just advanced statistics. As much as models are a black box, we understand the inputs and optimizations that...
I think I hold 2 salient points that back up a lot of your thoughts:
ML is pretty much just advanced statistics. As much as models are a black box, we understand the inputs and optimizations that change the outputs. It seems like one more tool to add to the data processing toolbox. I think the association with AI is just to drive hype, which leads to point 2...
Capitalism ruins most things. Spotify's and Youtube's algorithms don't push us to content we'll enjoy or help us grow, they push us to using the platform more. For all of these new tools - LLMs, Stable Diffusion, etc - they could be helpful tools as well but the common denominator is we need to sell and grow as much as possible in a capitalist system.
This is a good and entertaining video. There are multiple themes in this (quite long) video, but one of them seems to be criticism of the technology itself. I have a slightly different take on...
This is a good and entertaining video.
There are multiple themes in this (quite long) video, but one of them seems to be criticism of the technology itself. I have a slightly different take on this part than the author (Eddy Burback). Most of the problems he talks about are not about technology per se, but in the lack of control of these tech companies. They are too large, too powerful, too amoral, and most of the world's governments are either captured by them or don't understand the technology well enough to regulate it properly.'
"We have a saying. 'Move fast and break things.' The idea is that if you never break anything, you're probably not moving fast enough." - Mark Zuckerberg
Only an asshole who doesn't have proper accountability for the damage he causes would say something so irresponsible. But this is the kind of person who is shaping the future of the world.
There is another layer of problems caused by AI that Eddy didn't cover: It is yet another tech bubble that is in the process of popping. As usual, the salespeople promise a Wizard of Oz, gullible people invest in it, then sooner or later it's revealed that it's just a dude behind a curtain pulling levers, and a lot of people lose money.
Is there more context to that Zuckerberg quote? In my personal experience, this saying usually applies to breaking the things you’re building, not all the stuff around you. Another way of putting...
Is there more context to that Zuckerberg quote? In my personal experience, this saying usually applies to breaking the things you’re building, not all the stuff around you. Another way of putting it would be that it’s worth breaking the product we’re building for a little bit every once in a while if it means we’re developing the product quickly. I don’t think it means if we’re not destroying things around us, we’re not going hard enough.
It’s also possible that I’ve been interpreting this phrase incorrectly for my entire professional life
It was the motto of an immature tech company. He was referencing facebook's internal systems; ie their core competency, but that was before Facebook was a platform that over a billion people used...
It was the motto of an immature tech company. He was referencing facebook's internal systems; ie their core competency, but that was before Facebook was a platform that over a billion people used regularly, some of them for their livelihoods.
It's ok to break an app regularly when people are just using it to share vacation photos. Facebook now gets used for e-commerce, for marketing, for meetings, for critical business processes, and a lot of people use it as a vital, and sometimes only communication tool. It's no longer economically or morally ok to break Facebook.
It was also extrapolated, at the time, to a lot of businesses in technology in a kind of cargo cult deference that is unfortunately common when relating technology to what they're doing in the FAANG world.
It turns out, most things that people use are important. That's why they use them. Yes, people spend some time on frivolous things like entertainment or mindless distraction, but most of the things people pay for, they pay for because they have at least some degree of importance to them.
I don't want my doctor's office to "move fast and break things". I don't want my car manufacturer, the people who make my phone and computer, my internet service provider, the companies who make the food I eat, the police who protect me, the corporations who build the airplanes I fly in to follow that mindset either.
In ok with maybe like 1% of the things I pay for to sacrifice their stability in pursuit of new features or services. It's become a philosophy that wayyy too many organizations tried to replicate, and it's no longer even appropriate in the company that he was talking about in context.
The quote is from before the current AI bonanza so it isn't directly related to AI. My point is that it is part of a Silicon Valley mindset that they take high risks without worrying about...
The quote is from before the current AI bonanza so it isn't directly related to AI. My point is that it is part of a Silicon Valley mindset that they take high risks without worrying about consequences. Moving fast and breaking things doesn't have to literally mean "Oops we have a css bug in the Facebook UI". It also obviously means that Zuckerberg has no problem with creating a product that has been really bad for civilization. It is bad for your mental health to read all the outrage and see all the ads for unnecessary products. It has undermined politics throughout the world by being a prime source of disinformation. It has been used to facilitate genocide. Facebook could mitigate these problems by taking proper responsibility for who advertises there and having proper moderation. But they simply don't give a shit about that stuff.
This is not the best source ever to support what I said, but please enjoy this episode of "Behind the Bastards" which discusses the topic: https://www.youtube.com/watch?v=NlSDeNkcgcM
I don’t really disagree with any of this. I think the examples you listed here are far better than a pretty standard tech startup mindset about iterating quickly on a product. I agree with your...
I don’t really disagree with any of this. I think the examples you listed here are far better than a pretty standard tech startup mindset about iterating quickly on a product. I agree with your characterization of Zuckerberg’s lack of accountability, I just wouldn’t use that quote as evidence of it. I don’t mean to be pedantic, I think it’s an important distinction.
I saved myself a lot of time by having AI summarize the comments for me, take that as you will: The comments on the article discuss the pros and cons of artificial intelligence (AI). Critics argue...
I saved myself a lot of time by having AI summarize the comments for me, take that as you will:
The comments on the article discuss the pros and cons of artificial intelligence (AI). Critics argue that AI is overhyped and misused, while supporters emphasize its potential benefits. There is ongoing debate about AI's impact on society, ethics, and the job market.
I'll be honest, I'm just kind of sick of these takes. Don't misunderstand, please - I also think AI is making the world worse in the long run. I do however think that the layman - and I'm quite comfortable calling Eddy Burback a layman on this subject - is simplifying and confusing a few things.
First, though, a quick glossary - I'll call marketed solutions like Claude, ChatGPT, Gemini, Midjourney and DALL-E "AI", and the underlying technology "machine learning" or "ML", so we can have this conversation without mixing up what torment nexus the capitalists are parading around with the actual research of actual scientists. And to give you a TLDR: AI bad, ML neutral but interesting.
I don't think machine learning is a bad technology. There's certainly an argument to be had about what ethical data collection for training purposes looks like, but I am neither knowledgable enough on the topic of copyright nor do I personally care to litigate this issue - but saying that there's "no application for any of this" and just calling the models and technology "bad" because the output isn't very high fidelity is just lazy. I've been following (and tinkering with) diffusion and transformer models since the mid 2010s. I personally was quite impressed when ChatGPT was released. GPT 3.5 and its contemporaries were incredibly competent compared to other solutions available at the time, and I was hopeful to see what could be done with them. Sure enough, a revolution was sparked. Suddenly, everyone worked on a cool ML project. Auto-GPT comes to mind, and the stuff that Stability AI was doing on the front of open-source image diffusion was really interesting to watch and keep up with. Seemed like there was a new breakthrough every other week.
The capitalists of the world however didn't look at a Stable Diffusion or an Auto-GPT and thought "whoa that's really neat, this teaches us so much about how computers and maybe even our own brain works, this is obviously not ready for prime time nor will it be in the forseeable future but just making computers do this at all is insane". They looked at a mediocre copywriting tool and thought that they can sell this.
And sell they did. I don't have to look far to find corporations including "AI" in products where it doesn't even make a lick of sense - that AI grill from CES comes to mind, and Google even replaced Assistant with a new AI powered version that struggles to set your pizza timer, but does recommend you put glue on it. This is failure, obviously, and the execs didn't understand the technology enough to consider this was a bad idea.
I'm not saying that Gemini is bad because it tells you to glue your cheese on. I'm saying that Gemini and its contemporaries are actually very good at the things they are intended for. This includes sentiment analysis, summarisation, that sort of thing. An example here: Kagi's Universal Summariser is an excellent tool, but it's not good at deterministic tasks like mathematics or counting the number of Rs in the word "Strawberry". And yet, that's what we see when people dunk on ML. "Look at this stupid idiot, can't even count right" you will hear in the echochambers of especially artists - more on that later - as if that was ever the point of the invention.
I think AI is a mistake. I don't think ML is a mistake. I think the technology should have been left to us nerds, people who know what they're doing and know the limitations and quirks of this (in this iteration) very new and unexplored frontier in computer science. And to that end, I don't think Eddy is even that wrong in the video - to quote Theo Katzman, I don't really want to sing along if a computer wrote the fucking song.
But who would even make a computer write a song with the intention of it becoming a hit? Hint: It's the guys with the dollar signs in their eyes. The technology itself is in my opinion completely neutral. Machine learning will not take your jobs, and machine learning does not steal from artists. People do. The people that choose to abuse these research efforts for their own personal gain, laziness or convenience. That's the problem. Not the computers calculating matrices.
I'm not sure I understand your point. On one hand, you agree that the current iteration of AI as marketed towards end consumers sucks. So what issue do you take exactly with these takes? That they are all too similar?
Don't get me wrong, I don't disagree with you at all. I think ML has a ton of credible uses, and I also agree that right now, beyond tech demos, it doesn't really offer any valid use to the end consumer. I think it can be used in varying ways to enhance products that are used by end consumers, but I really don't understand what AI offers to you in its current state.
There is one exception; character art for TTRPG characters. Most of these are small groups of 3-6 people and commissioning art of your character is something that you do only when money is no issue or you have spent a particularly long time with a character. Paying for good art is expensive, as it should be, you're paying for someone else's creativity.
While the way that current AI image generators are trained is media piracy, I don't consider this specific usecase too problematic, because the overwhelming majority of people generating art fo their TTRPG characters never pay an artist to make them "real" anyway.
What I find entertaining about the current situation and the hype behind AI is less about the technology itself, and how an army of business and marketing MBAs are fevereshly working to sell you a product that is obviously flawed and has a miniscule amount of any real usecases, and how the best they come up with is an app that improves business communications when the issue I have with a lot of these communications is that this email didn't need to be better worded, this email didn't need to happen at all.
Or the Gemini ad about a father having the AI compose a letter for his daughter to send to an athlete. Like that's what you come up with? Who made this? One of the most endearing things a parent can do; sitting down to compose a letter with their child for their greatest idol, and your idea is to abstract it into a fucking chat prompt? Who made this? What room of, likely creatives working in the advertisement industry, thought this sounds like a good idea? Was there even one parent in the room?
AI advertisement presents everyone in it as a clueless idiot who can't decide on anything. It is, at it's core, the ad industry doing it's best to sell you a shit product. Paradoxically, AI is also a huge bubble right now, and more people are throwing more money at it than anything else, because modern free market capitalism needs it's quarterly growth or it's fucked, apparently.
Yes, maybe I wasn’t clear. Sorry about that. My issue with these takes (and let’s be honest, Eddy isn’t anywhere near the worst offender) of “AI output bad therefore AI bad” is that it’s narrow-minded and does nothing to address the real problems with AI. The hordes of none-to-underpaid Indian workers on Mturk classifying images. The moderators that have to sift through all the heinous shit online. But no, AI is bad because it’s “annoying” and “steals from artists”. It almost pisses me off. Like, the Waymo segment. Do you realise how incredible it is that this tech works at all? But no, let’s reject it because it makes known exploiters Uber and Lift obsolete and because my Waymo didn’t understand navigating a parking lot. There’s good and valid criticism of both AI and ML, but this ain’t it.
I would like to stake my claim that it does steal from artists. Training an ai to create works of art based on others peoples works of art without their consent or fair recompense is stealing. You know that if the situation was reversed, these companies pulling the bs wouldnt think twice to litigate in response. If i took all of microsofts ip and ran it through ai image gen to make and sell my own products, im sure Microsoft would call the lawyers when they found out. But i digress because that is up to interpretation.
What i find not to be up to interpretation is that it is stealing jobs from artists already. Big companies are already using it to do art for promotional material. See this debacle with microsoft. Thats an artist that lost a gig, or even an employed position at another company. Here is a video about a professional graphic design artist getting layed off after find out his company had been training an image gen AI on HIS art and work for months or years. He was literally replaced by an AI that is designed to copy his style. Those are just the first two examples off the top of my head, but this going to get worse. Publicly traded companies will take the path of least resistance, and paying a licensing fee is far cheaper than employing a real human being.
I also find your defense of waymo to be a bit dismissive of the workers it would replace. Known exploiters they may be, Uber and Lyft did create jobs for a lot of people that were otherwise struggling to get anything. It may not be a good job, or a fairly structured job, but its better than nothing. Real people still see real benefit from being employed (contract workers, not proper employment) through them. Nobody sees any benefit from waymo except the people at the head of the company. You are taking these jobs away from already vulnerable and exploited people. I am electronics-pilled guy, and no matter how many times i get impressed that waymo cars are real and functional, its always swallowed by the fact that in the end, the only meaningful thing these cars change is putting another human out of work.
Whenever I come across this line of reasoning it makes me think of how I've read that people were very skeptical of automobiles in general around the end of the 19th century, due at least in part to how it would put horse-drawn carriage drivers and repairmen out of work.
I'm of the opinion that jobs becoming obsolete due to the advancement of technology should not be used as a reason to rally against or artificially impede that advancement. Instead we should be thinking about how to provide other opportunities to or otherwise help those people who are out of work, ideally moving towards an eventual post-scarcity society where nobody needs to work to live, and instead everyone can use their time to pursue the things that they actually want to be doing. We'll never get there if we prioritize outdated human-powered technologies over better automated alternatives just so people can stay locked into jobs that they would (in many cases I'd guess) rather not be doing anyway if they didn't have to.
I'm sorry to say that I just disagree, even if I'm pedantic about it. Taking copies of someone's work to train a model on is not theft, it's copyright infringement. Theft is depriving someone of a physical good, and the originals still exist in this case. The reason why I care to differentiate is simple: Obviously I get what you're saying, you're saying that it's as bad as theft and that's fine, we can probably agree on that, but saying it is theft then creates a disconnect - if it's so bad, why do you have to pretend it's something else? Dr. Stuart Ashen said this quite well: You can think Athlete's Foot is as bad as having an ear infection, but if you say that Athlete's Foot is an ear infection you're just going to confuse people. Same with this. If you want to litigate whether or not this practice should be illegal, you need to make your case within the current vernacular of the truth, for lack of a better term.
As for your other two points, I don't think you understood what I wanted to say. I'm sorry that happened to Nadestraight, but that just sort of proves my point. It was the corporation that betrayed him and copied his work without permission, not the AI model. Humans made that decision, not a machine. This is, to the company, functionally no different than outsourcing an IT department to India because it's a fraction of the cost as the in-house guys that are then laid off. Nothing new, and not the fault of a new technology (I'm going to go out on a limb and say that you probably wouldn't campaign for oursourced Indian labour to be illegal) - Again, AI vs ML in my original post. Same with Waymo. Originally, it was a research project, and now it's a product, because humans decided it can be sold. I don't think I can fault the technology for that - and for what it's worth, I see this more as a "changing of the guard" than a death knell for Uber and Lift. The taxi lobby kicked and screamed through that transition 10 years ago as well, and yet it happened. That's to be expected in capitalism, so if we agree that that torment nexus is here to stay, we also have to shrug and accept this change.
This current implementation sure. But there is the pipedream that we'd have a proper city wide/country wide/continent wide network of cars which could provide as good if not better alternative to private car ownership.
Especially in sparsely populated areas where it doesn't make sense to have a local Uber/taxi network this would provide an equalizer compared to big cities.
Of course I wouldn't trust Google to do that but there's potential beyond just replacing drivers.
We already have an as good if not better alternative to car ownership, and have for hundreds of years. Public transit.
A fleet of robo cars that shuttle people to and from where they want to go sounds great, because you're imagining a fictional best case scenario where everything works perfectly. In real life, these things will be rolled out as soon as the massive corporations creating them decide they're good enough to be profitable, which means they'll be as expensive, dangerous, inconvenient, and annoying as the market can bear. As evidenced by online advertising, social media, crypto currency, ride sharing, and every other "disruption" innovation of the past 20 years, that's quite a lot.
Local and state governments could, right now, change policies to incentivize transit and pedestrian oriented development, stop subsidizing private car ownership as much as they do, and start subsidizing walkability and public transit, and within a very short time, people's lives would be improved via tried and tested technology that's existed for hundreds of years while tackling the biggest contribution to climate change. We don't do that, not because the technology isn't there, but because private companies can't get as rich off of it.
That's the most frustrating thing about transportation when people talk about it as if it's not a problem that's already been solved in many places.
Unfortunately, with the incentives currently in place, robo cars will only ever take off if huge companies discover they can make an absolutely ridiculous amount of money off of it, which pretty much precludes it from ever being a good deal for the average person. Even if alphabet's intentions with waymo are as pure as can be, the auto industry is such an entrenched and valued part of the American economy that they will spend every last lobbying dollar they have to ensure they get their cut in order to "defend American jobs" or whatever the justification is, and we're going to be right back at square one.
The problem of private car ownership has never been technological, it's always been political.
Yep, I use public transit too and in addition own a car. It makes everything way more convenient as I don't live in the city center so everything takes time with the switches etc. I'm not going to explain more than that about my personal life because it's nobody's business but I'm pretty sure lots of people can relate to using and having both options.
And I'm aware that big companies are for profit that's why I mentioned that I probably wouldn't trust Google on this.
But a man can dream of a world where robotaxis would be implemented as a complimentary service for public transit.
Where do you draw the line between end consumers and enhancing products used by end consumers?
I pay for ChatGPT and apart from various programming tasks I've used it to write a semi-scientific paper, where it transformed my notes part by part into written chapters that needed a lot of editing, but it cut my time by about 50% (I was still paid for 100% of the time). This seems like a specialized end user application.
The first professional early adopters of Midjourney were concept artists who often drew over its output because their creations don't get published as is anyway. I recently used Dall-E 3 to create a cool vintage logo to be used as an intro for my band's live video recording. I had to work on it for about an hour to fix spelling and typography errors because Dall-E is a bit outdated now, but it's creative and I couldn't get the same level of creativity out of Flux, which does text perfectly. I suspect Midjourney would do both, but I didn't want to pay for it - ideally the process would be just to take it and use it as it is, since it's only shown for like 10 seconds anyway, and I think we may be at that point already. That seems like end consumer usage.
Last week I talked to a guy who works for a company that offers finetuned translation LLMs adapted for specific subjects. Imo that is one area that is clearly not a bubble (and that is going to have bad consequences for an already underpaid field).
AI assisted search engines I guess are a typical example of enhancing an existing product, but some of the results are so novel that they pretty much are new products, like LLMs searching through released scientific papers and giving reasonably accurate summaries with citations to specific papers, because this lowers the barrier to learn something in a subject where you lack formal education enough that it allows more people to do it. I use this and it has been very practical for me.
And the "non-AI" applications of machine learning are even broader: for the live music recording mentioned above we used a tool that splits a stereo recording into individual channels, so that you can do a mix of something that wasn't recorded in multichannel. It now works well enough that it's actually useful, although imperfect, and it singlehandedly saved that recording for us. I used an open-source tool and most online services are just an application of this same open-source tool (but paid, that is undoubtedly a bubble), but some are not and they will probably continue making money because it's simply useful for musicians not just for recording but also for learning new songs, for people who do remixes and mashups or for karaoke.
Rereading the paragraph you're quoting, I didn't phrase myself clearly. In the same vein as the person I was responding to, I divided AI (as in, chat gpt and others) and ML into two broad categories to keep it simple. I agree that ML has quite a few uses. I friend of mine uses the audio splitter for songs to create mashups a lot, and I helped them set it up because they're not too great with computers and running python through cmd was difficult for them.
I've not had luck with AI assisted search engines so far. The famous pizza glue incident or that one time Google's Gemini recommended suicide come to mind.
The core of Burback's criticism remains, though. Where is all this efficiency going? Not into our pockets, that's for sure. You think if those concept artists become more efficient, they're suddenly going to get paid better? As freelancers, maybe. Worst case it's going to lead to an expectation from employers and soon enough you'll need to use it to keep up with demands.
I'm happy that chat gpt works for you, I personally cannot use it at work because legally it's unclear who code would belong to. I've never found the writing too compelling because there is no purpose behind it, and I like writing more than editing. Even if I'm more efficient, I'd rather write it myself. And that's what I mean: I'm not interested in watching over a robot doing stuff. I would rather do stuff.
This is mildly off-topic, but
I recently started using perplexity.ai for searching through papers. It sometimes gets the summary wrong, but it's better than google or bing and it almost always gives me a solid place to start (I always have to check those papers anyway). I use the free version, so feel free to try it if you ever need something like that. Pizza glue etc. imo all come from companies releasing significantly rushed products when trying to outcompete OpenAI.
Recommending Ultimate Vocal Remover, it's a free GUI for common models, it's pretty good, doesn't have any external dependencies and downloads the models through its own download tool.
It depends, for me writing an article was just a job, and one of the parts I dislike the most is the beginning of an article or beginning of a chapter, the introduction and planning how to structure the information from my notes that I want to include there. ChatGPT completely removed this obstacle and even if I had to rewrite 75% of what it gave me (most of the time it was less), it was faster and more pleasant for me to finish it. Not discounting that this is a subjective thing, of course.
There's a bizzare trend of advertisements that insult the consumer going back for practically forever. There are yogurt commercials that pretend that men are idiots. There have been coffee commercials that say that a man will leave a wife if she doesn't make coffee good enough. It's basically like pickup artist negging. As much as people want positive things, the thing that really makes people motivated to do things is the possiblity of losing something. So as a result many ads try to capitalize on FOMO.
I think I hold 2 salient points that back up a lot of your thoughts:
ML is pretty much just advanced statistics. As much as models are a black box, we understand the inputs and optimizations that change the outputs. It seems like one more tool to add to the data processing toolbox. I think the association with AI is just to drive hype, which leads to point 2...
Capitalism ruins most things. Spotify's and Youtube's algorithms don't push us to content we'll enjoy or help us grow, they push us to using the platform more. For all of these new tools - LLMs, Stable Diffusion, etc - they could be helpful tools as well but the common denominator is we need to sell and grow as much as possible in a capitalist system.
A great video reckoning with the current advertised uses of AI to end consumers, how it works in reality and where we go from here.
This is a good and entertaining video.
There are multiple themes in this (quite long) video, but one of them seems to be criticism of the technology itself. I have a slightly different take on this part than the author (Eddy Burback). Most of the problems he talks about are not about technology per se, but in the lack of control of these tech companies. They are too large, too powerful, too amoral, and most of the world's governments are either captured by them or don't understand the technology well enough to regulate it properly.'
"We have a saying. 'Move fast and break things.' The idea is that if you never break anything, you're probably not moving fast enough." - Mark Zuckerberg
Only an asshole who doesn't have proper accountability for the damage he causes would say something so irresponsible. But this is the kind of person who is shaping the future of the world.
There is another layer of problems caused by AI that Eddy didn't cover: It is yet another tech bubble that is in the process of popping. As usual, the salespeople promise a Wizard of Oz, gullible people invest in it, then sooner or later it's revealed that it's just a dude behind a curtain pulling levers, and a lot of people lose money.
Is there more context to that Zuckerberg quote? In my personal experience, this saying usually applies to breaking the things you’re building, not all the stuff around you. Another way of putting it would be that it’s worth breaking the product we’re building for a little bit every once in a while if it means we’re developing the product quickly. I don’t think it means if we’re not destroying things around us, we’re not going hard enough.
It’s also possible that I’ve been interpreting this phrase incorrectly for my entire professional life
I agree. "Move fast and break things" is about updating facebook.com. Not about releasing problematic AI products into the world.
It was the motto of an immature tech company. He was referencing facebook's internal systems; ie their core competency, but that was before Facebook was a platform that over a billion people used regularly, some of them for their livelihoods.
It's ok to break an app regularly when people are just using it to share vacation photos. Facebook now gets used for e-commerce, for marketing, for meetings, for critical business processes, and a lot of people use it as a vital, and sometimes only communication tool. It's no longer economically or morally ok to break Facebook.
It was also extrapolated, at the time, to a lot of businesses in technology in a kind of cargo cult deference that is unfortunately common when relating technology to what they're doing in the FAANG world.
It turns out, most things that people use are important. That's why they use them. Yes, people spend some time on frivolous things like entertainment or mindless distraction, but most of the things people pay for, they pay for because they have at least some degree of importance to them.
I don't want my doctor's office to "move fast and break things". I don't want my car manufacturer, the people who make my phone and computer, my internet service provider, the companies who make the food I eat, the police who protect me, the corporations who build the airplanes I fly in to follow that mindset either.
In ok with maybe like 1% of the things I pay for to sacrifice their stability in pursuit of new features or services. It's become a philosophy that wayyy too many organizations tried to replicate, and it's no longer even appropriate in the company that he was talking about in context.
The quote is from before the current AI bonanza so it isn't directly related to AI. My point is that it is part of a Silicon Valley mindset that they take high risks without worrying about consequences. Moving fast and breaking things doesn't have to literally mean "Oops we have a css bug in the Facebook UI". It also obviously means that Zuckerberg has no problem with creating a product that has been really bad for civilization. It is bad for your mental health to read all the outrage and see all the ads for unnecessary products. It has undermined politics throughout the world by being a prime source of disinformation. It has been used to facilitate genocide. Facebook could mitigate these problems by taking proper responsibility for who advertises there and having proper moderation. But they simply don't give a shit about that stuff.
This is not the best source ever to support what I said, but please enjoy this episode of "Behind the Bastards" which discusses the topic: https://www.youtube.com/watch?v=NlSDeNkcgcM
I don’t really disagree with any of this. I think the examples you listed here are far better than a pretty standard tech startup mindset about iterating quickly on a product. I agree with your characterization of Zuckerberg’s lack of accountability, I just wouldn’t use that quote as evidence of it. I don’t mean to be pedantic, I think it’s an important distinction.
This looks like an AI generated text, irony!
I saved myself a lot of time by having AI summarize the comments for me, take that as you will:
The comments on the article discuss the pros and cons of artificial intelligence (AI). Critics argue that AI is overhyped and misused, while supporters emphasize its potential benefits. There is ongoing debate about AI's impact on society, ethics, and the job market.