What Did OpenAI Do This Week? - 11/06/2023 [+30 LINKS]
OPENAI SEES FIRST-OF-ITS-KIND LIBEL CASE
Another busy week for Sam Altman as he buttered up Presidents and PMs, asked students in Delhi to amaze and e-mail him, and called in China to help make AI safe, but the happy tour had a sour ending. Why? OpenAI was served a first-of-its-kind libel lawsuit for allegedly damaging the reputation of a Georgia radio host named Mark Walters.
Fred Riehl, a journalist for gun news website AmmoLand.com, asked ChatGPT to provide a summary of the case The Second Amendment Foundation v. Robert Ferguson. ChatGPT is alleged to have stated that the suit was a legal complaint filed by the founder and executive vice president of the Second Amendment Foundation (SAF) against Mark Walters, host of Armed American Radio. ChatGPT identified Mark as SAF’s treasurer and chief financial officer and explained that he was accused of embezzling money from SAF. However, everything ChatGPT stated was wrong. Whilst Walters is strongly aligned with SAF’s Second Amendment activism and has spoken at their events, he holds no position in the group and has never been their treasurer or CFO. Walters had become the subject of an LLM “hallucination” (and AI falsehood). When Riehl then asked ChatGPT to further provide him with the exact passage of the lawsuit mentioning Walters, the chatbot allegedly provided detailed text referencing “Defendant Mark Walters (‘Walters’)” that does not exist in the actual complaint. It has also been alleged that ChatGPT went beyond getting the facts wrong to create fake documentation that supports its lies, ‘right down to a phoney case number’. And so, Mark Walters filed a first-of-its-kind libel lawsuit against ChatGPT for allegedly damaging his reputation.
So far, responses to the case are divided due to its unprecedented nature. As Mack DeGeurin explains, ‘many internet companies have avoided libel suits in the past thanks to the legal shield of Section 230 of the Communications Decency Act, but the protections are unlikely to apply to chatbots because they generate their own new strings of information rather than resurface comments from another human user’. His discussion with Eugene Volokh (UCLA professor and the author of a forthcoming law journal article on legal liability over AI models) makes for a compelling read given the implications of this case could have a lasting influence on how we treat AI hallucinations. Here is an excerpt below
SO WHAT?
On Friday, Miles Klee asked two legal experts (not ChatGPT!) if Walter has a case. Several holes were pointed out. From asking can ‘dignitary harms’ really be claimed when the material was only shown to one person (Riehl), who didn’t believe it?’ to Walters complaint doesn’t include the exact prompt that Riehl originally fed into ChatGPT, leaving open the possibility that the reporter had somehow steered it toward making up the story’. All up, legal analysis (to date) would suggest Walter may not be a first libel-case winner vs. OpenAI. Eugene Volokh (UCLA professor and the author of a forthcoming law journal article on legal liability over AI models) ‘stressed the specific limitations of this case don’t necessarily mean other libel cases couldn’t succeed against tech companies down the line’. The jury is yet to decide on whether tech companies can realistically be found liable for the content their bots produce. Whatever decision is made sets a huge precedent for more than just AI companies.
Could RAG and Contextual AI be an answer if OpenAI’s answer to removing hallucinations proves to be elusive, and libel cases become a threat to how typical LLMs operate? Maybe. Contextual AI (tagged as “AI That Minds Your Business”) launched out of stealth with $20 million in seed funding. Its aim is to build the “next generation” of LLMs for enterprise. Whilst at Meta, Douwe Kiela (co-founder with Amanpreet Singh, who worked at AI startup Hugging Face) led research into a technique called retrieval augmented generation (RAG), which forms the basis of Contextual AI’s text-generating AI technology. RAG augments LLMs with external sources, like files and webpages. Given a prompt, RAG will look for data within the sources that might be relevant. Then, it packages the results with the original prompt and feeds it to an LLM, generating a “context-aware” response (e.g., “The answer is X, according to Y”). That’s everything that didn’t happen when Fred Riehl asked ChatGPT to provide him with a summary of The Second Amendment Foundation v. Robert Ferguson. Conversely, (as Fred experienced and Kyle Wiggers pointed out) a typical LLM (e.g. ChatGPT) ordinarily returns answers up to a certain date and fails to cite the source, and it’s architected in a way that makes it difficult to remove — or even revise — their knowledge. One thing is clear, Contextual AI is certainly a company to watch.
WANT TO INNOVATE LIKE OPENAI? Order your copy of the second volume of ‘Disruptive Technologies’ now!
OpenAI was sued for libel by Mark Walters, a radio host from Georgia after ChatGPT claimed he embezzled money. /Gizmodo
Sam Altman privately reassured developers that OpenAI won't compete with them beyond ChatGPT. /Business Insider
Sam Altman told a conference in Beijing (via video link) that OpenAI planned to ‘open-source more of its models in the future, as part of its efforts to drive AI safety’. /Business Standard
Sam Altman said there should be no regulation on smaller companies, at a conference in India's New Delhi. /Reuters
OpenAI announced its hiring engineers/researchers ‘who do rigorous and thoughtful work understanding and evaluating LLMs’. /@barret_zoph
Sam Altman said OpenAI has ‘no plans to go public any time soon’ at a conference in Abu Dhabi. /ReutersSam Altman spoke by phone with Israeli Prime Minister Benjamin Netanyahu. /@IsraeliPM
Greg Brockman announced GPT models are soon to be made available to U.S. federal agencies using Microsoft’s Azure cloud service. /@gdb
GPT-5 confirmed as still ‘not being trained’ by Sam Altman. /Tech Crunch
ChatGPT adopted by City of Yokosuka after favourable trial results. /Japan Times
ChatGPT-4 announced @Carbon Health aimed at reducing a doctor’s workload by generating records and billing codes. /@gdb
ChatGPT on iOS improved iPad support and Shortcuts integration. /The Verge
Sam Altman visited Microsoft Corp's R&D centre in Israel and predicted a "huge role" for Israel in reducing risks from artificial intelligence. /Reuters
Sam Altman told 100 South Korean startups ‘that any regulatory framework has got to make sure that the benefits of this technology come to the world.’ /Reuters
Sam Altman encouraged South Korean President Yoon Suk Yeol, to supply chips in AI boom. /Reuters
Sam Altman funded Lab-grown meat maker Uncommon in a $30 million raise. /Reuters
Sam Altman said ‘it would be a mistake to put heavy regulation on the field right now or to try to slow down the incredible innovation’ in front of an audience at Tel Aviv University. /Reuters
Sam Altman said that AI can improve government services and deserves backing at a New Delhi event hosted by India’s Economic Times. /CommunicationsToday
Sam Altman said he is willing to invest in Korean startups and collaborate with bigger chipmakers like Samsung Electronics. / KoreaJoongAngDaily
Sam Altman and Atty Eleti (a software engineer at OpenAI) invited university students in Delhi to ‘amaze them’ and email sam@openai.com to gain “developer access immediately” and/or get a job. /Business Standard
GPT-4 is being pirated by people scraping exposed API keys. /Vice
ChatGPT needs to go to college. Will OpenAI pay? /Bloomberg
Microsoft and Google inserted ads into AI experiments without giving brands an opt-out. /Reuters
Zoom launched AI summaries of the meetings you’ve missed. /TheVerge
LinkedIn launched a generative AI tool to write ad copy. /Venture Beat
Rishi Sunak proposed a research centre, global summit, and regulatory body for artificial intelligence in Britain in his meeting with Joe Biden. /Spectator + /gov.uk
Contextual AI launched from stealth with $20m in funding to build enterprise-focused language models. /TechCrunch
Cohere (an AI foundation model company that competes with Microsoft-backed OpenAI) raised $270 million in funding. /Reuters
Zuckerberg shared AI products to reassure employees of Meta’s strategy. /Techcircle.