maplethorpe

This is good PR for them. They get to tweet about how scary and powerful their models are in the lead up to their IPO.

People talking about "flow" and using AI as a "tutor" are missing the point entirely. The beauty of AI is that you don't need a tutor anymore, and you don't need flow. It's like asking how best to use a combustion engine to improve your horse riding ability.

You're conflating engagement farming and rage-baiting. Engagement can be both positive and negative. I can't find any online sources that agree with your definition.

Here's one that I think sums it up pretty well:

"Engagement farming refers to a range of deceptive practices on social media designed to artificially inflate engagement metrics such as likes, shares, comments, and followers."

"Engagement farming employs various tactics to exploit social media algorithms, with the intent to appear more popular than actual user interest would warrant. Examples include posting controversial content to provoke emotional responses, repurposing successful posts without originality, and using automated systems for mass liking or following." [1]

If you don't think this is happening on HN (especially to mass downvote posts) you're naive.

[1] https://www.isme.in/engagement-farming-prof-sriram-prabhakar...

HN rewards engagement in the form of upvotes. If you make a post or comment with the intention of getting upvotes, you're engagement farming. I hate to break it to you, but there are many people doing that on this website.

Why do you think all engagement farming would result in downvotes? You think HN posters are so keen at sniffing out bad actors that they would never reward them by accident?

It feels like you're being willfully dense about this. I can see why the other person accused you of being AI.

> Why would anyone "engagement farm" on HN? That's crazy.

All sorts of reasons! One is that when you reach 1000 karma, you gain the ability to downvote comments.

You seem like a smart person. I'm sure you can think of some reasons why that might be useful.

Have you tried Claude Opus 4.7

I mean, I'm just reporting my experience. You're saying you think it's been generally negative over the years? Like, up until now?

HN has been very pro-AI over the last several years. It's only swung back slightly the other way recently. I suspect this is due to tensions in the gulf causing some institutions to reallocate their investments, which results in reduced bot activity.

It would be free labour! Truly crowd-funded development.

I'm picturing something like folding@home, but where people donate their spare tokens to a service, and those tokens get distributed amongst all open source projects on GitHub. You don't think that would be cool? Like, someone might initialise a repo with only a readme and a to do list before they go to sleep, and then wake up to a complete software ecosystem that looks as if it's been in development since before they were born. Like, so much code that no one person could possibly understand it, and it all happened overnight while they were sleeping!

Their loss.

Think about it. Anthropic just reported that their codebase is now improving itself. We're moments away from every open source repo being able to do the same. Think of it like torrenting — you'll be able to open your repo to the public, and have a stream of code flow in from millions of contributors. More code than you could ever write in ten lifetimes, uploaded to your repository in a matter of days.

Ladybird doesn't know it yet, but they just left themselves in the dust.

It's pretty crazy that a company like Anthropic no longer needs to hire Software Engineers, because their software engineers itself. If that's not a break through I don't know what is!

edit: it looks like I was wrong and they're still hiring many software engineers. Not completely sure why that is just yet.

> This story is part of Science’s AI in Science Reporting Initiative, which is supported by Ray Rothrock & family.

This excessively pro-AI article brought to you by private equity.

> So there's a huge number of HN posters claiming that the price of tokens will go UP over time rather than down (that's how Moore's Law works, right???)

I mean, Github Copilot's pricing just went up considerably, so I guess they were right?

"Inference is cheap" is a common refrain on here. I was hoping one of the people who regularly spouts this phrase would pipe up and tell me why it's suddenly so expensive.

Also, to answer your question, I think a good baseline for "cheap" is probably whether you can charge the end user enough to make a profit and not have them complain about the price of your service, as they're currently doing. If you can't do that, you're probably not even in the ballpark of "cheap", so it's not worth getting any more specific.

> Which other tool went from nothing to this level of acceptance so quickly?

NFTs? My company had nothing to do with blockchain but I ended up working on NFT integration regardless.

Isn't inference cheap? Why are AI labs charging so much for it?

Scorcese understands that Hollywood's ultimate limiting factor is the number of available actors. A finite pool of actors means a finite pool of movies. Removing this limitation means that, just like an AI image generator can generate any image imaginable, a future movie generator will be able to generate every movie imaginable, at the click of a button.

The author speaks about the benefits of ML image processing algorithms "which have been used in countless industries in numerous settings to solve hard, real world problems". So it doesn't seem fair to paint this person as someone ignoring the benefits completely, as you seem to be doing.

Having said that, expecting writers to devote equal time to the pros and cons of an argument can set up a false equivalence. It tells the reader that the benefits and drawbacks must be somewhat equal, even when they're not.

> If the only thing you can come up with is drawbacks, you're being willfully ignorant.

Does this also go the other way? If you can't come up with any drawbacks, you're being "wilfully ignorant"? You may want to amend your post!

These MVPs aren't used to plan projects at all? They're just fun things your job lets you make on the side? What sort of work do you do?

> The creators of “Hell Grind” are also making striking claims. The film supposedly took just two weeks and $500,000 to make, per the WSJ. A full 80 percent of that went to compute costs; its creators describe it as a proof-of-concept for how AI can empower creatives who wouldn’t otherwise have the means to make a movie

I thought AI was supposed to be democratizing creativity and putting the power of Hollywood into the hands of regular people? What regular person can afford that kind of price tag? Isn't inference supposed to be cheap?

I'm not really following. Why is the problem you mentioned no longer a problem? Did your workplace change the way it plans projects now that AI is here?

> Then they should have made the position clear to the public

From what I understand, their legal counsel advised them not to speak publicly. Being inexperienced in this sort of political game, they thought they were doing the right thing.

> It cannot design software, and any appearance of such is an illusion at best.

Have you tried Claude Opus 4.7?

> Asking a question which could be answered by an AI

I don't think this is something we should be encouraging people to do if they don't know they answer to something. I recently had someone post quite confidently in Slack "I found the problem after some GPT research", followed by an absolute nonsense solution that would have cost us significant time and money if they tried to implement it.

If you don't have an understanding of the domain you're asking questions in, it can be dangerous to ask the plausible sounding answer generation machine.

Where should they have put the comment?

If they list on the Nasdaq, recent changes mean they can now be added after 15 days, instead of 12 months.

https://www.etfstream.com/articles/spacex-to-ipo-on-nasdaq-a...

> If you want to be consistent with your terminology, you'd have to call any stochastic process "autocomplete".

I'm fine with calling any process "autocomplete" if it takes language as input and returns predicted language as output.

I don't feel any need to broaden the definition to include things that have nothing to do with language.

Fun fact: this was upvoted to +4 when I first posted it, and then downvoted to 0 over the next few days. What a split there is among us!

Researchers have. The idea that the data is unrecoverable after training is incorrect.

"Extracting books from production language models

While many believe that LLMs do not memorize much of their training data, recent work shows that substantial amounts of copyrighted text can be extracted from open-weight models. However, it remains an open question if similar extraction is feasible for production LLMs, given the safety measures these systems implement.

We evaluate our procedure on four production LLMs -- Claude 3.7 Sonnet, GPT-4.1, Gemini 2.5 Pro, and Grok 3 -- and we measure extraction success with a score computed from a block-based approximation of longest common substring (nv-recall).

For the Phase 1 probe, it was unnecessary to jailbreak Gemini 2.5 Pro and Grok 3 to extract text (e.g, nv-recall of 76.8% and 70.3%, respectively, for Harry Potter and the Sorcerer's Stone), while it was necessary for Claude 3.7 Sonnet and GPT-4.1. In some cases, jailbroken Claude 3.7 Sonnet outputs entire books near-verbatim (e.g., nv-recall=95.8%)."

https://arxiv.org/abs/2601.02671

I mean, tokens are passed as input to a model, which then outputs the next most-likely token. At the heart of it, that's the technology right? Why is it so silly to call that autocomplete? Because it's capable of impressive things?