Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Anthropic says Alibaba illicitly extracted Claude AI model capabilities

reuters.com

Post reply on HN
Print view View original

931–940 of 1001 posts

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#931

Post by freejazz »

fny wrote: post #762
anon373839 wrote: post #380

Earlier quoted context omitted.

> These complaints of distillation are inflating the problem to make it sound worse than it is Unfortunately, the Reuters piece itself is complicit in this dramatization. The lede paragraph parrots Anthropic's talking point that distillation is an "attack", without using quotes that would alert the reader that this framing is a corporate talking point. Distillation is NOT an attack.

Distillation may not be an attack, but it is a ToS violation and could be seen as IP theft. Any reasonable company would be pissed if a competitor, especially at Ali Baba's size, leveraged that company's R&D to compete. It is in this sense, a corporate attack. If you want to roll your eyes at distillation concerns, you might need to excuse Anthropic for originally using pirated material to train their models.

> you might need to excuse Anthropic for originally using pirated material to train their models

You have it backwards

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#932

Post by Retr0id »

InkCanon wrote: post #617
someguyornotidk wrote: post #582

What exactly is illicit about what they did? Legally, model output cannot be protected by IP laws whether domestic or international. The most they can hope for is civil relief which is a stretch given the literally illicit methods they used to train their models. Ahtoropic got treated the same way it has been treating everyone else. This is the bed they made and now they, too, have to sleep in it.

Anthropic is master of Newspeak (see previously bugs -> vulnerabilities wrt Mythos). Distillation violates their terms of service, which is a civil offense, not a criminal one. It is not illicit, illegal nor breaks any laws.

It's a clever choice of words because "illicit" does not necessarily mean illegal, so they're technically not wrong, even though that's the connotation they clearly want to convey.

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#933

Post by xhinker2 »

If Anthropic’s accusation is substantiated — if using another model’s outputs to train your own model is considered “illicit extraction” — then everyone in the AI industry is guilty.

If you fine-tuned a model on GPT-4 outputs, you distilled GPT-4. If you used Claude to generate training data for your classifier, you distilled Claude. If you learned anything from any model’s outputs and used that to improve your own system or your own brain, you distilled it. The line between “learning” and “distilling” is non-existent. Intelligence is distillation. That’s literally how learning works — you expose yourself to high-quality outputs, internalize patterns, and generate your own.

If I use Anthropic’s model to learn and train my own brain, am I also distilling their model?

The accusation confuses learning with theft.

https://xhinker.medium.com/pot-calling-the-kettle-black-why-...

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#934

Post by takinola »

lysium wrote: post #411
yokisan wrote: post #224

Earlier quoted context omitted.

One would think Anthropic could point Mythos at this to solve the reseller problem outright: - Purchase multiple accounts via resellers - Send messages that contain a UID - Capture these in Anthropic's logs - Shut down account. Use any metadata to identify related accounts /loop

This only shuts down the account you have bought in the first, plus a few others if it is shared. > Use any metadata to identify related accounts How does that work? I think this is the most important part to have an impact on the „thousand“ bot accounts.

Rather than go after accounts, you’d go after the hosting facilities. The hosting companies are more likely to be able to identify related accounts so working with them will give more scalable results.

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#935

Post by 0xbadcafebee »

sorenjan wrote: post #708
0xbadcafebee wrote: post #11

There's two basic kinds of distillation: 1) the massive [and dumb] method where you ask a question and use the answer as reinforcement (Black Box), and 2) more targeted distillation where you use one model to directly inform/train/guide another model (RLAIF). The latter is basically fine-tuning the model with direction from another model. Thousands of businesses do this every day to fine-tune. This is almost certainl…

Doesn't "real" distillation use the logits instead of the final tokens? I would classify this more like using a model to generate synthetic training data.

Distillation is a category of techniques which generally speaking all extract knowledge from a target model to feed into a new model. Logit distillation requires access to the source model layers; final token distillation doesn't. The former is more effective, but the latter can be done with generation tokens alone.

This article explains the difference (and addresses "they're distillin' our models!!!"): https://dev.to/p0rt/how-model-distillation-actually-works-an...

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#936

Post by 0xbadcafebee »

killerstorm wrote: post #746
0xbadcafebee wrote: post #11

There's two basic kinds of distillation: 1) the massive [and dumb] method where you ask a question and use the answer as reinforcement (Black Box), and 2) more targeted distillation where you use one model to directly inform/train/guide another model (RLAIF). The latter is basically fine-tuning the model with direction from another model. Thousands of businesses do this every day to fine-tune. This is almost certainl…

I'm sorry, but you got the terminology exactly backwards. Training on the answer is called supervised fine-tuning. Just for the sake of clarity: 0. Full distillation uses logits of the teacher model - that's much more information than the text itself. This is a kind of distillation used inside labs, but one can't distill Claude this way as logits are not available via API. 1. Supervised fine-tuning on synthetic data…

I agree I left out option 0, but I think the other two were presented correctly?

- Black box distillation uses direct answers to questions and conversation style. This is less useful as you still have to do supervised fine-tuning on the answers, as they may be wrong, and don't lead to greater insights (which reinforcement learning does)

- RLIAF relies on preferences and values to judge answers. These don't need supervised fine-tuning and help guide the new model to better answers rather than just correcting specific previously asked answers

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#937

Post by Perepiska »

dualvariable wrote: post #236
gruez wrote: post #140

Earlier quoted context omitted.

>This is great for competition! Chinese vendors offering a cheaper solution = what economics told me the free market was all about. Yeah, like all those Chinese bootleggers selling DVDs for a few dollars rather than $20. Free market! https://news.ycombinator.com/item?id=48664814

"Information wants to be free" Anthropic profited from training its models on all kinds of copyrighted information, live by the sword, die by the sword... Their model weights, training data, training methods, etc are all going to leak to China over time. Nobody on a site named _Hacker_ news should be all that upset about this.

Los ricos también lloran (1979)

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#938

Post by Rudybega »

handoflixue wrote: post #533
mcintyre1994 wrote: post #446

Earlier quoted context omitted.

Supposedly the details of the ‘jailbreak’ are that you give it insecure code and say “fix this code”, and it does, and then you ask it for test scripts and that’s effectively an exploit against the unfixed code. If true then I have no idea how anyone’s going to release a useful model that doesn’t have the same jailbreak. https://www.theregister.com/security/2026/06/15/feds-freaked...

If that's the extent of the jailbreak, then the government should have banned every existing LLM - their story only makes sense if there's some Fable-specific capability that got unlocked.

I mean, it makes a lot of sense if the current administration operates in bad faith. I'll leave discerning that to you.

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#939

Post by wayeq »

someguyornotidk wrote: post #582

What exactly is illicit about what they did? Legally, model output cannot be protected by IP laws whether domestic or international. The most they can hope for is civil relief which is a stretch given the literally illicit methods they used to train their models. Ahtoropic got treated the same way it has been treating everyone else. This is the bed they made and now they, too, have to sleep in it.

> This is the bed they made and now they, too, have to sleep in it.

How will they sleep at night on that giant pile money.

Post reply on HN