Facebook Whistleblower Alleges Meta's AI Model Llama Was Used to Help DeepSeek (cbsnews.com) 10
A former Facebook employee/whistleblower alleges Meta's AI model Lllama was used to help DeepSeek.
The whistleblower — former Facebook director of global policy Sarah Wynn-Williams — testified before U.S. Senators on Wednesday. CBS News found this earlier response from Meta: In a statement last year on Llama, Meta spokesperson Andy Stone wrote, "The alleged role of a single and outdated version of an American open-source model is irrelevant when we know China is already investing over 1T to surpass the US technologically, and Chinese tech companies are releasing their own open AI models as fast, or faster, than US ones."
Wynn-Williams encouraged senators to continue investigating Meta's role in the development of artificial intelligence in China, as they continue their probe into the social media company founded by Zuckerberg. "The greatest trick Mark Zuckerberg ever pulled was wrapping the American flag around himself and calling himself a patriot and saying he didn't offer services in China, while he spent the last decade building an $18 billion business there," she said.
The testimony also left some of the lawmakers skeptical of Zuckerberg's commitment to free speech after the whistleblower also alleged Facebook worked "hand in glove" with the Chinese government to censor its platforms: In her almost seven years with the company, Wynn-Williams told the panel she witnessed the company provide "custom built censorship tools" for the Chinese Communist Party. She said a Chinese dissident living in the United States was removed from Facebook in 2017 after pressure from Chinese officials. Facebook said at the time it took action against the regime critic, Guo Wengui, for sharing someone else's personal information. Wynn-Williams described the use of a "virality counter" that flagged posts with over 10,000 views for review by a "chief editor," which Democratic Sen. Richard Blumenthal of Connecticut called "an Orwellian censor." These "virality counters" were used not only in Mainland China, but also in Hong Kong and Taiwan, according to Wynn-Williams's testimony.
Wynn-Williams also told senators Chinese officials could "potentially access" the data of American users.
The whistleblower — former Facebook director of global policy Sarah Wynn-Williams — testified before U.S. Senators on Wednesday. CBS News found this earlier response from Meta: In a statement last year on Llama, Meta spokesperson Andy Stone wrote, "The alleged role of a single and outdated version of an American open-source model is irrelevant when we know China is already investing over 1T to surpass the US technologically, and Chinese tech companies are releasing their own open AI models as fast, or faster, than US ones."
Wynn-Williams encouraged senators to continue investigating Meta's role in the development of artificial intelligence in China, as they continue their probe into the social media company founded by Zuckerberg. "The greatest trick Mark Zuckerberg ever pulled was wrapping the American flag around himself and calling himself a patriot and saying he didn't offer services in China, while he spent the last decade building an $18 billion business there," she said.
The testimony also left some of the lawmakers skeptical of Zuckerberg's commitment to free speech after the whistleblower also alleged Facebook worked "hand in glove" with the Chinese government to censor its platforms: In her almost seven years with the company, Wynn-Williams told the panel she witnessed the company provide "custom built censorship tools" for the Chinese Communist Party. She said a Chinese dissident living in the United States was removed from Facebook in 2017 after pressure from Chinese officials. Facebook said at the time it took action against the regime critic, Guo Wengui, for sharing someone else's personal information. Wynn-Williams described the use of a "virality counter" that flagged posts with over 10,000 views for review by a "chief editor," which Democratic Sen. Richard Blumenthal of Connecticut called "an Orwellian censor." These "virality counters" were used not only in Mainland China, but also in Hong Kong and Taiwan, according to Wynn-Williams's testimony.
Wynn-Williams also told senators Chinese officials could "potentially access" the data of American users.
Llama (Score:4, Informative)
That is the point of open source.
Fear mongering nonsense (Score:2)
Open source is good, much better than control by monopolists or governments
All of science, engineering, art and music benefit when ideas are shared
I have no doubts (Score:2)
DeepSeek used it as much as most other models that came after GPT-3.
I don't know if they actively created more data, but there are a lot of public datasets that are created using GPT-models and that are used by almost all models that came after the dataset. If someone puts a reasonable dataset online, you can assume that the large companies add it to their datasets.
Phrases coming from GPT are already known as GPTisms if you follow the creative writing (and rp) communities. Tell them that a shiver ran down y
Re: (Score:2)
Oops, I meant of course OpenAI. But for Llama the situation is certainly similar.
AI models are all results of incest, the open source community even uses methods to merge models (with the same base model) directly in the hope that they get the best traits of each of the merged models.
Re: I have no doubts (Score:2)
"The resulting story itself is just low quality"
Project, much?
Re: (Score:2)
Won't you consider a story low quality, that repeats the same 30 phrases over and over?
I've tried enough LLM to know their strengths and weaknesses. And good prose is not a strength of most of them.
OpenAI is building a model that should have no problem with that (They showed for an example short story with none of the usual slop), but I think they didn't even offer it on web and you can bet they won't release the weights (someone should sue them to remove the "Open" from their name).
OPSEC - Don't share private data (Score:4, Insightful)
I know FB can be fun. After all, which billions of other people don't want to see my live pictures of a cruise to the Bahamas, rob my house, steal my car from the airport parking lot, but don't forget to hit the Like button!!!
Facebook and IG and all the other "social media" companies are NOT governmental. They have a contract, which most people DON'T read but then click "I agree." Try reading it. It says they get to do whatever they want with your posted data. SOME of them also say they own your data in perpetuity.
In an Agreement there is an Offer and Satisfaction. So first, you Agreed to their Terms. Second, they offered you a place to ramble your meaningless life events and post pictures. Then they Satisfied their Offer by ensuring that what you posted is now available for the world to see.
AND IF YOU HAVE A PROBLEM with how THEY SHARE the data YOU GAVE THEM FREELY then go look in the mirror for the problem.
I'd write more, but I need to post this so I can link to it on FB so ALL MY FRIENDS will see HOW CLEVER I was on /. today.
It's a thing. Go check the mirror for the problem. Don't blame an autistic inventive dude. Mark Z never came to YOUR house and figured out what YOU were up to and posted it. YOU did.
Stop calling Llama open-source (Score:4, Insightful)
Llama has license restrictions that make it not-open-source.