post_below's recent activity
-
Comment on Patagonia sues drag queen Pattie Gonia for trademark infringement in ~lgbt
-
Comment on Patagonia sues drag queen Pattie Gonia for trademark infringement in ~lgbt
post_below Link ParentWithout more info it's a leap but it's interesting to think about. AI psychosis (or whatever we end up calling it) is a real thing. Those two sentences do smell like AI, but it's not enough text...Could it be that they've been using an LLM as an advisor without realising the sycophancy, which may have led them to think they can get away with something as dumb and selfish as this?
Without more info it's a leap but it's interesting to think about. AI psychosis (or whatever we end up calling it) is a real thing. Those two sentences do smell like AI, but it's not enough text to really say one way or the other. It does seem unlikely that any human would advise them to handle the situation the way they have.
It's hard to say if it's dumb or just cynical. Maybe they expect to drum up financial support for their righteous battle and come out ahead despite having no chance of winning in court.
-
Comment on If you let AI do your writing, I will come to your house and kill you in ~tech
post_below Link ParentGood point, the longer the context window the weirder things get. And context trimming/automatic compaction confuse things even more.Good point, the longer the context window the weirder things get. And context trimming/automatic compaction confuse things even more.
-
Comment on If you let AI do your writing, I will come to your house and kill you in ~tech
post_below Link ParentDetails aside, yes the difference between two models can be dramatic, the technology is far from monolithic. The same is true for different harnesses, even different scaffolding. Experience with...Details aside, yes the difference between two models can be dramatic, the technology is far from monolithic.
The same is true for different harnesses, even different scaffolding. Experience with one model, in one interface, often doesn't generalize. Even with the same model, guidelines and other initial prompting make a huge difference so two people's experience with the same model can be very different.
-
Comment on Patagonia sues drag queen Pattie Gonia for trademark infringement in ~lgbt
post_below Link ParentOh wow they even copied the Patagonia logo and sold clothing under it.Oh wow they even copied the Patagonia logo and sold clothing under it.
-
Comment on If you let AI do your writing, I will come to your house and kill you in ~tech
post_below Link ParentYes definitely, but it sounds like you're using Gemini? In my limited experience with it, it's a distant third among the big 3 for most use cases. I'd recommend trying Chat GPT or Anthropic...Yes definitely, but it sounds like you're using Gemini? In my limited experience with it, it's a distant third among the big 3 for most use cases. I'd recommend trying Chat GPT or Anthropic models.
feeling like my instructions are being interpreted by a subtly vindictive genie in a bottle?
Hahaha... I love human writing
-
Comment on Patagonia sues drag queen Pattie Gonia for trademark infringement in ~lgbt
post_below LinkThe headline reads like a big company bullying someone, however: So Patagonia had to choose between losing their trademark and taking action, Pattie forced their hand by filing for a trademark,...- Exemplary
The headline reads like a big company bullying someone, however:
Wiley filed a trademark application in September to use the brand Pattie Gonia to sell clothing
So Patagonia had to choose between losing their trademark and taking action, Pattie forced their hand by filing for a trademark, essentially on their name, for purposes of directly competing with them. They're suing for $1 in damages.
“This is a betrayal of Patagonia’s core mission. Because if they’re ‘in business to save the home planet’, why are they suing a climate activist?” Wiley said.
Either they weren't listening when their lawyer explained how trademark law works or they're being willfully disengenuous.
In a statement, Patagonia told the Guardian: “Over the past several years, we’ve tried to find a path forward that would allow Pattie Gonia to continue their work while also protecting the Patagonia trademark. These conversations have included multiple proposals – each intended to support that path – along with ongoing dialogue and genuine efforts to avoid this ending up in court. Unfortunately, we could not reach an agreement.”
This tracks, I'd rank Patagonia among the most beneficial companies on earth. 98% of the company is owned by a non profit and 100% of company profits go to fighting climate change and other environmental causes (over $100 million annually).
The important point is that they let Pattie Gonia do their thing for years despite the pun because there wasn't a conflict that required them to defend their brand or risk losing it.
It's frustrating because it sounds like Pattie could have continued with their activism indefinitely if they'd just not filed for a trademark, and likely could continue even now if they withdrew the application and scrapped plans to sell clothing under the name Pattie Gonia. Maybe it's a publicity stunt?
-
Comment on If you let AI do your writing, I will come to your house and kill you in ~tech
post_below Link ParentI wish it was mundane. From my perspective it looks like a dramatic overhaul of content creation that will have a reverberating impact on culture and creativity for generations. I'd be willing to...I wish it was mundane. From my perspective it looks like a dramatic overhaul of content creation that will have a reverberating impact on culture and creativity for generations. I'd be willing to sacrifice quite a lot to stop all of the prose on the web (and elsewhere) from being increasingly replaced by slop.
Not just because I don't like reading it, somehow it hurts more to watch other people interacting with it as though they were interacting with a sentient being. Seeing people in comment threads defending and arguing about views expressed in an LLM generated article written by no one is deeply dystopian.
The OP piece, and others like it, aren't likely to make any difference in how it all plays out but I appreciate them anyway. We can't stop the flood but at least we can acknowledge it's happening.
-
Comment on If you let AI do your writing, I will come to your house and kill you in ~tech
post_below Link ParentWhile it won't fix any of the fundamental problems with LLM prose, a few sentences and some "do" and "don't" examples in claude.md (or whatever they call the initial prompt field in the web/app...While it won't fix any of the fundamental problems with LLM prose, a few sentences and some "do" and "don't" examples in claude.md (or whatever they call the initial prompt field in the web/app interface settings) can dramatically reduce those annoyances.
-
Comment on There's a hundred illegal erections in the hills behind my parents' house in ~hobbies
post_below LinkI knew nothing about this and otherwise likely never would have heard of it. This knowledge will probably serve no purpose in my life but I'm happy about it anyway. This is what makes the internet...I knew nothing about this and otherwise likely never would have heard of it. This knowledge will probably serve no purpose in my life but I'm happy about it anyway.
This is what makes the internet great, thanks for posting.
And also that journalist can fuck all the way off.
-
Comment on How I feel about LLM (AI) writing in ~tech
post_below Link ParentThe method is really interesting, solve the problem of fine tuning introducing LLMisms by running an extra step to move it back in the direction of the original distribution. It makes a lot of...The method is really interesting, solve the problem of fine tuning introducing LLMisms by running an extra step to move it back in the direction of the original distribution. It makes a lot of sense. There are downsides though.
I tried out the demo, asked it to write a blog post about LLMs because it seemed fitting. And you're right, if I'd come across the result in the wild I wouldn't have pegged the output as AI generated. However I wouldn't have read it either. It's written at a very low grade level and it makes false claims. It's considerably less accurate than modern LLMs. It accomplishes the goal of being hard to detect by making the hallucinations sound like ignorance and the rhetorical tells look like poor writing instead of AI giveaways.
At some point (maybe even soonish) LLM prose may indeed be impossible to tell apart from human prose, but at least for right now the majority of it is still pretty obvious.
-
Comment on I made my own Reddit alternative in ~tech
post_below LinkI really appreciate the pushback in this thread against low effort dismissals. OP made something they think is cool, engage with it in good faith or ignore it. That doesn't mean you have to like...- Exemplary
I really appreciate the pushback in this thread against low effort dismissals.
OP made something they think is cool, engage with it in good faith or ignore it. That doesn't mean you have to like it, but bring something besides "this is bad". What about it is bad? If you don't have something specific that means you should probably just ignore it
IMO it's bad "because AI" isn't enough. Not only because it's far too general, but because it's very clear that LLM based tool use is going to become more and more common no matter what we think about it. We need to figure out how we're collectively going to deal with that and "AI bad" isn't going to help us get there.
For example, you might ask the question: If this is mostly vibecoded, how do you know it's secure? How will people's PII and other data be protected? Or you could ask: If it's vibecoded, what's your plan for when you reach the end of the greenfield honeymoon and the complexity hits a level that's hard for coding agents to deal with? How will you maintain a mental model of the code if you're not writing it? Are you fluent in all of the languages (and possibly frameworks) used in the project? If not, will bugs that agents can't fix be unfixable?
@Staross Cheers for being up front about vibecoding. I've seen so many vibecoded projects where the author tried to pretend it was something else (and sometimes outright lied about it). The honesty is nice.
Good luck with the project! I watched a (non AI) video on your site about large pot blanching and learned something.
-
Comment on What do folks carry in their hiking/backpacking/camping first aid kits these days? in ~hobbies
post_below LinkI've been hiking and backpacking for a long time, in a lot of different conditions, and I've never encountered, or even heard of, a situation where someone was injured and the outcome was...I've been hiking and backpacking for a long time, in a lot of different conditions, and I've never encountered, or even heard of, a situation where someone was injured and the outcome was significantly affected by first aid supplies. Not saying it doesn't happen, but it's so rare that worrying about it seems like a waste of bandwidth.
The most important things to have with you are water purification and GPS. Offline mapping on your phone works great. Solar charging or a backup battery are nice.
If you're backpacking you're already carrying everything you need to survive for a long time if something goes wrong. If you're day hiking, and you don't fall off a cliff, you're rarely far from help.
One thing that's useful for overnights is the right clothes. Body temp is the next most important thing after water. Wool is magic, cotton is not your friend.
I suppose if you experience a lot of anxiety, feeling like you're prepared might be more important. And honestly in that case you don't want a splint or clotting agents, you want satellite communication and hiking buddies.
-
Comment on Startups in Berlin in ~finance
post_below LinkIt looks like there's some human involvement in this LLM written article.It looks like there's some human involvement in this LLM written article.
-
Comment on Bun has been rewritten in Rust in ~comp
post_below Link ParentThe thing about tests... When the model hits a failed test, or a parse error, or otherwise unexpected behavior, it starts making decisions, and if you've ever used a coding agent on anything...The thing about tests... When the model hits a failed test, or a parse error, or otherwise unexpected behavior, it starts making decisions, and if you've ever used a coding agent on anything non-boilerplate you know exactly what those decisions look like.
A lot of times they're ok, sometimes they're great, but sometimes they're incredibly bad. An extensive test suite and the code in another language can reduce that problem but not eliminate it. When you have as much code as they just vibed... I really don't envy the maintainers. There's going to be a strong incentive to slam the hood back down and have agents do the work going forward. Plus they're in Anthropic's world, everything is vibecoded, so there's a cultural incentive too. It may not be Claude Code yet, but it's already on the path.
Note that I'm not making any predictions, a year ago I wouldn't have bet on Claude Code being anywhere close to maintainble at this point and it's still going strong. It ain't pretty, but millions of people use it every day anyway.
-
Comment on Gemini 3.2 Flash rumored to hit 92% of GPT-5.5 performance at lower cost in ~tech
post_below Link ParentI just want to add, if you're going to buy a subscription for everyday use, get it from an actual model provider, not any of the countless 3rd parties that are basically resellers with...I just want to add, if you're going to buy a subscription for everyday use, get it from an actual model provider, not any of the countless 3rd parties that are basically resellers with questionable harness improvements. So that means: Anthropic, Chat GPT or Google. As R3qn65 said, any of those will be fine for basic use. Probably Chat GPT as it's more multi-modal (better at generating response in more formats, Anthropic models can't generate images for example).
-
Comment on Bun has been rewritten in Rust in ~comp
post_below Link ParentYou make some good points. Definitely using Rust was a good call. But I have to add, it will definitely be buggier, less idiomatic, less elegant and more bloated than before. That's just where LLM...You make some good points. Definitely using Rust was a good call. But I have to add, it will definitely be buggier, less idiomatic, less elegant and more bloated than before. That's just where LLM coding agent technology is right now. It's not as if being under the Anthropic umbrella gives them magical powers to make agents perform better. Look at how buggy, verbose and convoluted Claude Code is. It would be different if it was human led with AI assistance rather than just letting the agents churn... but it doesn't sound like that's how they did it given the short timeframe.
Caveat: if they used Mythos and it's lightyears ahead of Opus, maybe it's better than I'm imagining. I'd say that's somewhat unlikely.
Whatever the case, Claude Code and Codex, despite their issues, have proven that you can ship messy, vibe coded software and people will use it. Vibecoded doesn't mean the new Bun won't be successful.
-
Comment on Overworked AI agents turn "marxist" in ~tech
post_below LinkThis article (and the referenced experiment) frame the situation inaccurately and then explain why that framing is wrong. Kinda weird, why not just skip straight to the truth? Less engaging that...This article (and the referenced experiment) frame the situation inaccurately and then explain why that framing is wrong. Kinda weird, why not just skip straight to the truth? Less engaging that way I guess.
This result is just prompting, if you give the agent context that sits close to marxist language, you greatly increase the chance of marxist language. No anthropomorphizing necessary.
There is an attempt at substance though:
We know that agents are going to be doing more and more work in the real world for us, and we’re not going to be able to monitor everything they do,” Hall says. “We’re going to need to make sure agents don’t go rogue when they’re given different kinds of work.”
The current generation of agents are entirely stateless. There is no practical use case where the circumstances the experiment creates will happen because, largely due to context window challenges, the agent doing the second repitition isn't likely to be the "same" agent that did the first. It will happen in a new context window. There's no continuity. If you let the context window get too full, performance degrades.
If we imagine persistent memory being involved, as happens in many harnesses, agents would never write their "feelings" about the work to memory unless you told them to. They wouldn't write it to thinking traces or responses in the same session for that matter either. That behavior is so far outside of the training distribution that it would effectively never happen without intentional prompting.
They're essentially inventing an alignment issue that doesn't exist in any realistic scenario.
That may change in the future, but in that future it wouldn't be the models the experiment is testing, it would be much more sophisticated models with different guardrails. It's hard to understand the point of the experiment.
I'm not against doing experiments because "why not?". But this seems like a pretty weak attempt to generate LLM anthropomorphization clicks.
-
Comment on How I feel about LLM (AI) writing in ~tech
post_below Link ParentSouless is a great way to articulate it. I've often used the word hollow but I think I like that better. There's no substance at the core, even if the parts contain valid informationI don't really know how to describe it but the writing just feels soulless. Usually with writing, I can usually understand the writer's intent and their thought process as I go through a piece from top to bottom. But with AI writing, it just feels like words on a screen. I don't really know how to articulate this better haha.
Souless is a great way to articulate it. I've often used the word hollow but I think I like that better. There's no substance at the core, even if the parts contain valid information
-
Comment on How I feel about LLM (AI) writing in ~tech
post_below Link ParentThat's definitely a grey area, I personally wouldn't say there's anything wrong with using AI to translate a thoughtful comment into another language (or clean it up in the same language) but it...That's definitely a grey area, I personally wouldn't say there's anything wrong with using AI to translate a thoughtful comment into another language (or clean it up in the same language) but it would also be an easy way to justify outright generated content.
Small correction as this comes up a lot: During fine tuning various patterns emerge that are not represented in the training data in anywhere near the same degree. LLMs don't actually write like normal people much of the time.