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

941–950 of 1001 posts

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#941

Post by AJRF »

AJRF wrote: post #521

There is so much hot air and guff around AI, so please if you don't believe me verify yourself, but GLM 5.2 is "good enough" to replace Claude Code / Codex. No it's not frontier, but it's beyond that point that Opus 4.5 hit where people started to really depend on Claude Code around last November time. It's also a fraction of the cost of a Claude Code subscription especially when you account for how high the usage li…

Article just posted by the NYTimes on Z.ai (maker and operator of GLM 5.2) gaining ground in US: https://www.nytimes.com/2026/06/25/technology/zai-china-arti...

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#943

Post by comfysocks »

anon373839 wrote: post #380
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…

> 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.

Similarly, xAI “attacked” Anthropic and OpenAI to train grok.

https://opentools.ai/news/xai-trained-coding-models-claude-o...

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#945

Post by walrus01 »

marcusb wrote: post #719
itopaloglu83 wrote: post #614

Earlier quoted context omitted.

I thought Xerox demoed something they haven’t implemented yet, and Apple turned a mockup into a real GUI.

They implemented it. They just couldn't get out of their own way to successfully sell it (See "Dealer's of Lightning"[0]) 0 - https://www.amazon.com/Dealers-Lightning-Xerox-PARC-Computer...

It's not so much that they couldn't sell it, but adjusted for inflation to today's money, a well equipped Xerox Star workstation with reasonable amount of RAM, HDD ($20000+ in 1981 money), would be the equivalent of $70,000 in today's money. It was very much a high end system affordable only by a few niche users.

It's not clear to me how much it cost them to build the Star, but it might have been pretty close to that 20000 figure if they included all the R&D costs that went into it, and the fairly specialized high end hardware at the time like 20MB HDD. And it was being done in fairly low quantities so not much economy of scale.

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#947

Post by 4d4m »

tristanj wrote: post #25

Here's what is happening: Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices. They achieve this by reselling capacity from pooled Claude Max accounts, payments fraud, and also reselling the model output & reasoning chains to various Chinese labs. They are subsidizing model access in exchange for user logs and reasoning traces, which they then sell as training data, allowing the…

Thanks. This is a widely misunderstood part of the market that the frontier companies are either unaware of, or don't see as a threat. Consumers are flocking to this, its better than dealing with their limits, changes, and opaque pricing. Funny enough, frontier companies are literally training their users to seek alternatives, local models, and distilled models to meet their throughput needs.

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#948

Post by geye1234 »

p4coder wrote: post #416
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.

Agreed! I had to do a double take and check the URL. I thought I am reading a press release rather than actual reporting.

Once you see it, you can't unsee it:

https://www.paulgraham.com/submarine.html

And if you have more time:

https://dn790002.ca.archive.org/0/items/kupdf.net_bernays-et...

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#949

Post by pseudohadamard »

tristanj wrote: post #25

Here's what is happening: Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices. They achieve this by reselling capacity from pooled Claude Max accounts, payments fraud, and also reselling the model output & reasoning chains to various Chinese labs. They are subsidizing model access in exchange for user logs and reasoning traces, which they then sell as training data, allowing the…

"How dare someone rip off the stuff that we've stolen!".

Re: Anthropic says Alibaba illicitly extracted Claude AI model capabilities

#950

Post by ethbr1 »

csomar wrote: post #581
lysium wrote: post #411

Earlier quoted context omitted.

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.

They probably will route you through different accounts. So with a single account you should be able to hit hundreds-thousands of accounts.

Especially if you did this in a sting-like manner and used subtle compound UIDs (read: unique combinations of innocuous words) then banned every account found at a later time.
Post reply on HN