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

OpenAI says it has evidence DeepSeek used its model to train competitor

ft.com

Post reply on HN
Print view View original

91–100 of 1001 posts

Re: OpenAI says it has evidence DeepSeek used its model to train competitor

#91

Post by bhouston »

mring33621 wrote: post #65

We demand immediate government action to prevent these cheaper foreign AIs from taking jobs away from our great American AIs!

> We demand immediate government action to prevent these cheaper foreign AIs from taking jobs away from our great American AIs!

That is exactly what Microsoft and Sam Alman are asking for. And they will likely get it because Trump really likes protectionist governments policies.

Re: OpenAI says it has evidence DeepSeek used its model to train competitor

#92

Post by SilverBirch »

I think OpenAI is in a really weak position here. There are essentially two positions you can be in: You can be the agile new startup that can break the rules and move fast. That's what OpenAI used to be. Or you can be the big incumbent who is going to use your enormous resources to crush your opposition. That's Google & Microsoft here. For Microsoft to say "We're going to tie you up in lawsuits about the way you trained this model" would be perfectly expected and they can use that strategy because at any given time they have 1,000 lawyers and lobbyists hanging around waiting to do exactly that. But OpenAI can't do that. They don't have Google or Microsoft's legal teams or lobbyists or distribution channels. SO whilst it's funny that OpenAI are kind of trying to go down this road, this isn't actually a strategy that is going to work for them, they're still a minnow and they're going to get distracted and slowed down by this.

Re: OpenAI says it has evidence DeepSeek used its model to train competitor

#94

Post by throwup238 »

marricks wrote: post #36
semking wrote: post #21

This is absolutely hilarious! :) ClosedAI scraped human content without asking and they explained why this was acceptable... but when the outputs of their training corpus is scraped, it is THEIR dataset and this is NOT acceptable! Oh, the irony! :D I shared a few screenshots of DeepSeek answering using ChatGPT's output in yesterday's article! https://semking.com/deepseek-china-ai-model-breakthrough-sec...

Also, DeepSeek is allegedly... better? So saying they just copied ClosedAI isn't really sufficient of an answer. Seems to be just bluster because the US Govt would probably accept any excuse to ban it, see TikTok.

It’s not better. In most of my tests (C++/QT code) it just runs out of context before it can really do anything. And the output is very bad - it mashes together the header and cpp file. The reasoning output is fun to look at and occasionally useful though.

The max token output is only 8K (32K thinking tokens). O1 is 128k, which is far more useful, and it doesn’t get stuck like R1 does.

The hype around the DeepSeek release is insane and I’m starting to really doubt their numbers.

Re: OpenAI says it has evidence DeepSeek used its model to train competitor

#95

Post by bilekas »

> “It’s also extremely hard to rally a big talented research team to charge a new hill in the fog together,” he added. “This is the key to driving progress forward.”

Well I think DeepSeek releasing it open source and on an MIT license will rally the big talent. The open sourcing of a new technology has always driven progress in the past.

The last paragraph too is where OpenAi seems to be focusing their efforts..

> we engage in countermeasures to protect our IP, including a careful process for which frontier capabilities to include in released models ..

> ... we are working closely with the US government to best protect the most capable models from efforts by adversaries and competitors to take US technology.

So they'll go for getting DeepSeek banned like TikTok was now that a precedent has been set ?

Re: OpenAI says it has evidence DeepSeek used its model to train competitor

#97

Post by adamnemecek »

throwup238 wrote: post #94
marricks wrote: post #36

Earlier quoted context omitted.

Also, DeepSeek is allegedly... better? So saying they just copied ClosedAI isn't really sufficient of an answer. Seems to be just bluster because the US Govt would probably accept any excuse to ban it, see TikTok.

It’s not better. In most of my tests (C++/QT code) it just runs out of context before it can really do anything. And the output is very bad - it mashes together the header and cpp file. The reasoning output is fun to look at and occasionally useful though. The max token output is only 8K (32K thinking tokens). O1 is 128k, which is far more useful, and it doesn’t get stuck like R1 does. The hype around the DeepSeek re…

Thanks for saying this, I thought I was insane, DeepSeek is kinda bad. I guess it’s impressive all things considered but in absolute terms it’s not great.

Re: OpenAI says it has evidence DeepSeek used its model to train competitor

#98

Post by superkuh »

beAbU wrote: post #66
marricks wrote: post #36

Earlier quoted context omitted.

Also, DeepSeek is allegedly... better? So saying they just copied ClosedAI isn't really sufficient of an answer. Seems to be just bluster because the US Govt would probably accept any excuse to ban it, see TikTok.

How can they ban something thats open source that you can just run on your own hardware?

There was an executive order passed by the previous administration that make using anything with more than 10 billion parameters illegal and punishable by government force if done without authorization. Of course like most government regulations (even though this is not a regulation, it is an executive action) the point is not to stop the behavior but instead to create a system where everyone breaks the regulation constantly so that if anyone rocks the boat they can be indicted/charged and dealt with.

https://www.federalregister.gov/documents/2023/11/01/2023-24...

>(k) The term “dual-use foundation model” means an AI model that is trained on broad data; generally uses self-supervision; contains at least tens of billions of parameters; is applicable across a wide range of contexts; and that exhibits, or could be easily modified to exhibit, high levels of performance at tasks that pose a serious risk to security, national economic security, national public health or safety, or any combination of those matters, such as by: ...

Re: OpenAI says it has evidence DeepSeek used its model to train competitor

#100

Post by sho_hn »

pilooch wrote: post #74
semking wrote: post #21

This is absolutely hilarious! :) ClosedAI scraped human content without asking and they explained why this was acceptable... but when the outputs of their training corpus is scraped, it is THEIR dataset and this is NOT acceptable! Oh, the irony! :D I shared a few screenshots of DeepSeek answering using ChatGPT's output in yesterday's article! https://semking.com/deepseek-china-ai-model-breakthrough-sec...

Any ML based service with an API is basically a dataset builder for more ML. This has been known forever and is actually a useful "law" of ML-based systems.

Aye, this should be obvious even to non-technical folks. Much has been written about how LLMs regurgitate the data they were trained on. So if you're looking for data to train on, you can certainly extract it there.

Plus of course for people within the tech bubble, plenty of research results on the value of synthetically augmented and expanded training data that put the impact past just regurgitating source data.

This whole episode is a failure of reporting what to expect next and projecting running costs etc. most of all.

Post reply on HN