US gov just peeved that they weren’t the first to start removing truth and censoring big AI. Cos China truth gonna be diff from US truth. It’s like Tick Tock - soon to only present Truthspeak. Only TRUMP truth is real not XI truth
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
OpenAI Alleges DeepSeek Used Its Models for AI Training
- Thread starter MacRumors
- Start date
- Sort by reaction score
That sounds more like ocidental protenctionism.
If we american do, then thats inovation. If China do, then they copied! Also it put in risk our national security, lets ban! Or lets force then to sell us!
Hope somehow the Chinese blocks the giveaway of the TikTok To US. Looks ridiculous.
If this happens, every country with reasonable size should do the same and force selling slices of facebook and instagram because of “national security”.
Sorry, that's not quite what I was saying. I also wasn't criticizing the DeepSeek team. What they did is hugely impressive! What I meant was their costs were higher than the about $6 million reported, as covered in this comment: . That was just the cost for the final training of the current model.
The good thing is that going forward, others should be able to independently train for about that cost, but the actual cost for the model is considerably higher.
Also, as covered in their papers, part of their training / refining of the model used the outputs of other LLMs. I like what they're doing, I'm simply pointing out that the total costs were not just $6,000,000.
For the record, I also have no issues with OpenAI's approach. I don't think they should be complaining about this though.
Last edited:
OpenAI stole my IP to train their models. I hope DeepSeek wrecks them (but I doubt it). The more people who steal from OpenAI, the better. Schadenfreude is mine.
While there may be some reasons to question AI trainers use of copyrighted internet content to train their models, that is not at all the same thing as basically stealing an AI model, adding some additional code, and calling it something brand new and more powerful. That would be like buying a stock Ford Mustang, adding a faster fuel injector and saying "we've created a much faster car, for only (the cost of the new fuel injection system)." If this turns out to be true, it would explain why OpenAI spent hundreds of millions building its model more or less from scratch, and DeepThink was able to create a "better" model for 6 million.
I don't think they are making an effort to pay sites for data. Or the writers of books they pirated to train their AI. Or whatever else they stole.
At best they're paying some sites (reddit?) to gain monopoly access to new user generated data. Not because they think it would be fair to compensate, but because the only way they know to compete is by trying to gain a monopoly. See their doom and gloom predictions designed to make governments put financial barriers to entry for new "AI" companies in the name of "regulating the perils of AI".
The Chinese propaganda bots are out in full force today.
This is one instance where I am happy to see "Piracy" happen.
No, we just aren't blind to the fact the Sam Altman and his ilk are hypocrites.
My caving group (exploration and mapping) has 40+ years of mapping and trip data for certain cave systems that only we have been in. It was all posted for anyone to read, if you knew where to look, but we have taken it all offline since AI services started scrapping similar sites.
Now it lives completely off the internet, and will stay that way for the foreseeable future. Thanks Altman.
Last edited:
My AI brings all the bros to the yard
And they're like, it's better than yours
Damn right, it's better than yours
I can teach you, but I have to charge
And they're like, it's better than yours
Damn right, it's better than yours
I can teach you, but I have to charge
LOL, so OpenAI trained its model with l
When someone disagrees with you or has a different opinion, you call them bots. How nice!
Yeah. Feels like someone saw an opportunity to make money on new tech, so they simply made it expensive because they could not because there were no cheap option availableI read a neat article yesterday about how the Deepseek programmers used PTX instead of CUDA for their coding because CUDA (a higher level language than PTX) didn’t expose certain functions/methods the Deepseek developers used to improve performance.
DeepSeek's AI breakthrough bypasses industry-standard CUDA for some functions, uses Nvidia's assembly-like PTX programming instead
Dramatic optimizations do not come easy.www.tomshardware.com
.
While I am very much a novice here (those with more knowledge feel free to chime in with corrections), I think there is a lot of misunderstanding with what DeepSeek did, how they did it, and how much it cost.
Distillation is, as I understand it, pretty much using one model to help teach another model. It's done all over the industry, and while against the terms of service, isn't "stealing". Even if DeepSeek did what they are accused of here (and I'd argue that they almost certainly did), they didn't copy OpenAI's source code or anything like that, just used OpenAI's model outputs to help teach the DeepSeek model to be better. (I'd also argue that what OpenAI did by scraping the open web for content to train its model isn't stealing either, but I know reasonable people disagree about that).
The fact that DeepSeek pulled this off is a huge win for everyone who hasn't invested tons of money massive data center buildouts for AI. I'd even argue it's a win for NVIDIA long term (while obviously not great short term). The fact that they open sourced (well, "open weighted" to be technical) their model is shocking, but obviously also very good for everyone. To quote a Daring Fireball article from earlier this week:
As far as costs, DeepSeek says that they didn't build the model for $5.6m, they said the final training run cost $5.6m. Big difference.
Distillation is, as I understand it, pretty much using one model to help teach another model. It's done all over the industry, and while against the terms of service, isn't "stealing". Even if DeepSeek did what they are accused of here (and I'd argue that they almost certainly did), they didn't copy OpenAI's source code or anything like that, just used OpenAI's model outputs to help teach the DeepSeek model to be better. (I'd also argue that what OpenAI did by scraping the open web for content to train its model isn't stealing either, but I know reasonable people disagree about that).
The fact that DeepSeek pulled this off is a huge win for everyone who hasn't invested tons of money massive data center buildouts for AI. I'd even argue it's a win for NVIDIA long term (while obviously not great short term). The fact that they open sourced (well, "open weighted" to be technical) their model is shocking, but obviously also very good for everyone. To quote a Daring Fireball article from earlier this week:
As far as costs, DeepSeek says that they didn't build the model for $5.6m, they said the final training run cost $5.6m. Big difference.
Towards the end of Tiananmen movement, the situation was chaos and large part of the movement was like Brownian motion. It was not like the initial demonstration by students in higher education sectors. This was what I heard from the number three on CPC's wanted list for 8964. According to him, youngsters (students or freshly graduates or just young ppl) gathered at the square were talking romance and having sex every night. Sorta like a lesser rave party. Anyway those deceased didn't deserve bullets.
Now go to ask ChatGPT about Israel's genocide in Gaza. You'll be told of lies after lies. I guess the moral of the story is censorship not that different around the globe.
Don't want to be hypocritical.
I use AIs from tiem to time, but given how most of these AIs scaped the web with very little concern towards copyright...
I cannot feel bad for Open AI.
Despite having an RTX card, it's about time nvidia lost some market evaluation.
It really started to become an unstoppable monopoly.
I use AIs from tiem to time, but given how most of these AIs scaped the web with very little concern towards copyright...
I cannot feel bad for Open AI.
Despite having an RTX card, it's about time nvidia lost some market evaluation.
It really started to become an unstoppable monopoly.
Reminds me of an old favorite album of mine from Edwin McCain.
It was his hit debut album after being discovered and pushed into the mainstream by Darius Rucker and Hootie & The Blowfish
music.youtube.com
It was his hit debut album after being discovered and pushed into the mainstream by Darius Rucker and Hootie & The Blowfish
Honor Among Thieves
Honor Among Thieves is Edwin McCain's first major-label album, released on CD and cassette tape on August 15, 1995 by Lava Records. Songs from Thieves also appear on McCain's live concert DVD, Tinsel & Tap Shoes.
Last edited:
How is that any different from OpenAI using user data ?
It’s worked for China really well. Their electric cars from BYD start at $19k. They can hurt worldwide car companies at that price. Tesla is failing in China now. China just copies but at an incredible value. Chinese phones can do 95% of flagship phones for half or more off.
User data is not proprietary and you sign it away when you sign up for the app.
Open AI is mad that China AI puts them out of a job. How does that feel Sam?