OpenAI Board Asked Anthropic For Merger

0:00Today on the AI Breakdown, OpenAI's board approached Anthropik about a merger over the weekend. Before that on the brief, while we were all distracted, Meta disbanded its responsible AI team. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information about our YouTube, our Discord, and our newsletter.

0:24Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five minutes. And yes, today, friends, OpenAI has finally slowed down enough to actually warrant a brief. My assumption is that lawyers are finally involved or something because it's gone remarkably quiet. But today we are actually catching up on some of the news that has been buried behind the scenes as everyone has been focused over there in that crazy San Francisco story. And one of those stories that has been buried, and it seems maybe was intentionally placed during this time of chaos, was the fact that Meta has disbanded its team that was focused on responsible AI.

1:02So this came from the information initially, as so much breaking AI news does. And this team was called the RAI. It was the Responsible AI team, and it was Meta's dedicated group that was focused on questions of guardrails, on AI safety. Indeed, Meta's so-called pillars of Responsible AI include accountability, transparency, safety, privacy, and more. Now, a representative from Meta said that the company will, quote, continue to prioritize and invest in safe and responsible AI development. And he also added that this team was not being let go, but were just being moved to other departments within the company.

1:33The representative said that those members will, quote, continue to support relevant cross-Meta efforts on responsible AI development and use. Now, one thing to note is that it was not exactly as though this team was a powerhouse inside Meta already. The team had been around since 2019, but during layoffs earlier this year, Business Insider said that those restructuring efforts had left the RAI to be, quote, a shell of a team. That report also suggested that our AI had very little autonomy, that it didn't have the ability to push its initiatives through. And so the question is how much it was just some face-saving lip service type of thing.

2:06Now, I think to the extent that one cares about responsible AI development in general, not just this team, the more important question for Meta is not so much whether this team needed to stick around, but how they're going to make sure that all of their efforts actually have this sort of mindset. Indeed, to give them the benefit of the doubt, is it possible that having these team members inside the generative AI product team, which is the core beating heart of the company's efforts in generative AI, might actually make them have more, not less influence? Who knows, and the company hasn't even tried to say that's the case.

2:38So ultimately, all we have to go on is the fact that this responsible AI team is being disbanded right as this major question of controversy around approaches to AI safety swirls up and sits in Silicon Valley. Now, to be fair, Jan LeCun, the chief AI scientist at Meta, has also used the chaos at OpenAI to reinforce how that company views things differently than, in particular, the AI safety-focused folks that seem to be one side of the OpenAI question. Jan tweeted yesterday, There is at least one industry research lab where the leadership believes that superhuman-level AI is attainable. Is a scientific research question not just a question of more compute and more data?

3:15Is not just around the corner, it will take a while. Is not an existential risk. Requires contributions from the entire research community because no one has a monopoly on good ideas, hence requires open source platforms and open research because no one has a monopoly on good ideas, and finally is going to change human condition for the better. I would suggest that so far after this whole fiasco, this position is more, not less in favor when it comes to how the technology industry is viewing the field. However, speaking of AI doomers, apparently some of them were on attend at a very secretive summit focused on AI security that was held last week in the mountains in Utah.

3:52VentureBeat writer Sharon Goldman says, The day before OpenAI CEO Sam Altman was fired, I attended a secretive AI security summit in Utah. I hung out with doomers, heard sobering predictions, and chatted with government officials. In hindsight, it helped me understand the OpenAI drama. So this event was hosted by ScaleAI. It was an invitation-only event, and in fact, Scale AI footed the bill for most of the attendees, although many journalistic organizations like Bloomberg paid their own way. Now, the thing that was interesting about this event is that it brought together people from the technology space as well as from the military establishment.

4:24Even though a handful of press were invited, everyone had to agree to Chatham House Rules, where they weren't allowed to report on exactly who said what. The topics were things like AI security risk mitigation, the AI roadmap for the U.S. cybersecurity and infrastructure security agency, and of course, AI competition between the US and China. According to Sharon, as well as the other reporters who were there, this was definitely a sober conversation that showed the extent to which the military and political establishment is taking the rise of AI very, very seriously. Now, interestingly, the two hosts of the event, ScaleAI CEO Alexander Wang and former GitHub CEO Nat Friedman, were both approached over the weekend by the OpenAI board to take on the role of new CEO.

5:03But that's something we'll get into in the main part of the episode. Moving over to Europe, there has been a bunch of bluster around how the EU is planning to implement its AI Act, and one of the big remaining questions has to do with their approach to foundation models. France, Germany, and Italy have just released a paper where they seem to have come to some amount of agreement on this question, and it appears that the focus is self-regulation and the principle that what should be regulated is not the technology itself, but the application of the technology. Now, there has been a big question here around this foundation model question.

5:36You have to remember that when it comes to the EUAI Act, this was a bill that was written in large part before the rise of ChatGPT and the emergence of this new generation of generative AI tools. Now, I think quite smartly, they punted many of the issues that have arisen in the last year, choosing to focus on the stuff that they had had time to actually debate. However, one of the things that got rolled in was this question of foundation models. A couple of weeks ago, France had started to mount a strong opposition to any regulation on foundation models which had gained support from these countries, Germany and Italy.

6:06They had asked Spain, who currently holds presidency of the EU Council, to retreat from the approaches that had been articulated in the past. However, in response to that, European Parliament officials walked out of a meeting to make it clear that to them, leaving foundation models entirely out of the law was not going to be an acceptable course of action. A new paper from France, Germany and Italy, or as Euractiv calls it, a non-paper, quote, shows little room for compromise. The paper basically says we need to take a self-regulatory approach and learn before we implement full rules. I imagine there is going to be a lot more debate around this, but it's significant enough that it actually jeopardizes the entirety of the EU AI Act.

6:41I think this might turn into a full episode sooner rather than later, so we will leave it there for now. Now, by the way, it is not just the EU that is struggling to figure out how to regulate generative AI. An FDA commissioner in the US has said that everyone is struggling to figure out how to regulate generative AI. With refreshing candor, Food and Drug Administration Commissioner Robert Califf said, the thing I think we are all struggling with is generative AI, which is a new thing and something we don't really know how to regulate at this point. Now, perhaps a tool for them will be some of the new classes that have just been launched by Amazon.

7:12Amazon becomes the latest company to launch an education initiative around AI, this time with a goal to educate 2 million people on these new tools. And to do so, they've launched eight free AI classes. Five of them are focused on developers, and three are focused on non-technical use cases specifically for business people. The new program is called AI Ready, and that 2 million person ambition is between now and 2025. Now of course, when it comes to Amazon and their approach to AI, what we're all really waiting to see is whether these rumors about them training a model that's bigger than GPT-4, that has been codenamed Olympus, is actually true.

7:46For that, however, we will have to wait and see, and so with that, we will wrap the brief. Up next, the main AI breakdown. Hello friends, just a quick note before we get to the main episode. As I mentioned yesterday, we are running an experiment in December. It is an AI educational and learning community. So basically what this means is that every day I'm going to post new content to this group. It will include short video tutorials, case studies, as well as some shared challenges that we'll do together. The idea is to experiment with different ways to actually help people learn the tools that are shaping the AI revolution and hopefully advance your journey to integrating AI into your projects, professions, and lives.

8:23This is a paid beta experience. it will be$20 for the month, and there are a limited number of slots available. It's going to be a really exciting journey. I appreciate all of you who have signed up so far. And if you are interested in participating, go to bit.ly slash AI beta. That's bit.ly slash AI beta. Hope to see you there. And now, back to the show. Welcome back to the AI Breakdown. Today, we continue our coverage of the saga at OpenAI. And I will start off by noting that things have taken a turn for the quiet. It is distinctly less loud and seemingly fast-moving than it had been over the past 72 hours or so, which to me suggests that some combination of lawyers and comms professionals got involved.

9:08But of course, that doesn't mean there aren't still really important updates to be had, including some very interesting intrigue around the possible outcome of a merger with Anthropic and OpenAI. Now, where we left the story yesterday was that after former Twitch CEO Emmett Shear had been announced as the new CEO of OpenAI, and after Microsoft CEO Satya Nadella had said that Sam and Greg were coming to lead a new division within Microsoft, well, that morning, a group of OpenAI employees said, absolutely not, this is ridiculous. At the time of recording yesterday, something like 500 out of 770 employees at OpenAI had signed a letter saying basically that they had no confidence in the board, that the board had not taken the opportunity to actually explain their actions, and that should they not resign and reinstate Greg and Sam, that they would be leaving to go to Microsoft with those two.

9:56Now, it turns out that that very impressive 500 plus out of 770 sort of number was only at 500 because the rest of the team was asleep. By the end of the day yesterday, it was over 745 of 770 employees that had signed the letter. As the information put it, the vast majority of OpenAI employees asked the board to resign. That includes 22 of the other 24 leaders of the company, according to the information's org charts, and one of those was Ian Hathaway, who's a partner at the OpenAI Startup Fund, which is a different legal entity than OpenAI. Now, the other who hadn't signed the letter, Jan Leakey, is a notable absence, given that he's the head of Artificial Intelligence Alignment.

10:35However, he also said in a post on X that he thought the OpenAI board should resign as well. So, update one is that there is near unanimity among OpenAI employees that things should not go down as they went down. The next new bit of information that we got was that Emmett Shear, the former CEO of Twitch, was not necessarily the OpenAI board's first choice to be the new interim CEO of the company. Now, if you missed the detail around why they were looking for another interim CEO, it's because the interim CEO that they had appointed in former CTO, Mira Murati, had gone against the board and was trying to use her new authority to bring Sam and Greg back.

11:11At that point then, the board got very, very quiet and started to go into full recruitment mode it appears. And once again, the information has dug up who they asked to join before they got to Emmett. One person they asked was Nat Friedman. Nat was the former CEO of GitHub, and he is also a prolific AI investor. The other candidate they asked before Emmett was Alex Wang, who's the co-founder and CEO of ScaleAI, who, as you heard about in this morning's brief, just held an AI security summit in the hills in Utah. Both of those two politely declined the offer. That left Emmett Shear, and almost immediately as he had joined, there were challenges.

11:44First of all, news outlets started digging up controversial tweets that he had sent in the past around all manners of topics. But beyond that, it was unbelievably clear that he had very little to no support from the employee base. According to people with knowledge of the matter, the vast majority of employees at OpenAI refused to attend an emergency all-hands meeting scheduled with Emmett Shear, and in fact, instead responded to the announcement in their Slack channel with an F-U emoji. Influential investor Vinod Khosla wrote, Time for Emmett Shear to resign before he becomes the only employee of OpenAI.

12:15Time to DM him, everyone. Now, given that there had been such an outpouring of support from the team at OpenAI, it wasn't surprising to learn that behind the scenes Sam was still working the channels and that the deal to move to Microsoft wasn't done. At around 1.30pm Eastern Time yesterday, Altman tweeted, We have more unity and commitment and focus than ever before. We are going to work together some way or the other and I'm so excited. One team, one mission. So what is actually going on behind the scenes? Well, according to The Verge, Altman, former President Brockman, and the company's investors are still trying to find a graceful exit for the board, say multiple sources with direct knowledge of the situation.

12:49About an hour later, Sam tweeted, Satya and my top priority remains to ensure OpenAI continues to thrive. We are committed to fully providing continuity of operations to our partners and customers. The OpenAI slash Microsoft partnership makes this very doable. However, as The Verge points out, quote, it's not clear how going to Microsoft with over 700 former OpenAI employees is compatible with ensuring OpenAI continues to thrive, or how that can reasonably be set as a priority for those former employees once they are working at Microsoft. Still reiterating the message, Microsoft CTO Kevin Scott said, to my partners at OpenAI, we have seen your petition and appreciate your desire potentially to join Sam Altman at Microsoft's new AI research lab.

13:27Know that if needed, you have a role at Microsoft that matches your compensation and advances our collective mission. A nice touch both to say it publicly and also to make sure that we're clear that there is matching compensation. Now, maybe the most interesting detail wasn't that they had asked people like Nat Friedman and ScaleAI's Alex to come be the CEO before Emmett Shear, but the fact that they had also explored an even more dramatic shift, which would have been a merger with Anthropic. Now, the background of this, of course, is that Anthropic was in many ways a spinoff of OpenAI. It was a group of people who had been at OpenAI who didn't like the way that they were approaching some of the safety issues in guardrails, and so who wanted to start their own lab focused on responsibility.

14:05Anthropic is also chock-a-block full of effective altruists, which is why, of course, Sam Bagman-Fried was one of their biggest investors. Now again, while it doesn't seem like there was any real consideration given to this, the fact that they had even discussed it shows just how off-the-rails things have gotten. These are two arch-rival companies, the two leading startup labs, and frankly, until the board's decision on Friday, Anthropic was in the distinct number two position. Now, had they actually pursued this, it would have caused a whole new set of messes, given OpenAI's relationship with Microsoft and Anthropik's relationship with Google and Amazon.

14:35But ultimately, it was not to be. The one other big interesting thing was that Microsoft CEO Satya Nadella did a round of press. Some of the highlights, as summed up by Krish Kastanova, were 1. Nadella said that they have all the IP and legal stuff to succeed with Sam and the team if they join Microsoft. 2. And maybe most significantly, even Satya still didn't know why the board fired Sam Altman. Now, this is wild to comprehend. We're sitting here banging our heads against the wall because the board hasn't been willing to tell the world what happened, but it sounds like they're not even telling the key actors behind the scenes.

15:07I talked about this yesterday on a podcast with Dimitri Kofinas that was published on the Hidden Forces podcast with Dimitri Kofinas, but this is just an insane state of affairs. Every time that I've seen someone try to say that we should give the board the benefit of the doubt, and that maybe Sam was up to much more than it seemed, I keep coming back to the fact that they haven't been willing to justify their actions. And in the absence of justifying their actions, everyone is going to assume the worst. Which in this case is just that it was some petty power struggle, that it wasn't actually about any big safety disagreement at all.

15:35They haven't even come out and said something to the effect of, it's too dangerous to tell you why, so you just have to trust us. They've literally said nothing. Instead, the only thing we've gotten is the big reversal from Ilya, which would seem to suggest that it isn't about some major safety issue, given that one would think that if that was the cause, that he wouldn't have been so willing to flip his position. And so where this leaves us is a continual liminal in-between. Forbes editor Alex Conrad tweeted, Founders working with OpenAI Models, how are you feeling right now? Joe Vazani says, Ripping it out before APIs start failing.

16:05Nathan Binok writes, Uneasy. Avni writes, Frustrated that Monday morning isn't about implementing end-of-year plans, but instead running contingency planning. Angry for how this impacts hundreds of incredible people. Sad that all this magic was so carelessly decimated or at least disrupted, with seemingly such little thought or explanation. Investor Manal Hassan writes, Most of my portcos are going to diversify and use a mix of LLMs going forward. From a business risk and fiduciary risk standpoint, this is the best route. Now, if that's the anecdotal part of the story, the evidence-based part of the story is there as well.

16:34According to the information, more than 100 OpenAI customers contacted Anthropic over the weekend. Others reached out to Google Cloud and Cohere. And this is the sort of thing that doesn't go away quickly even if Sam does come back. What's more, those companies are taking advantage of the situation. Anthropic announced this morning that Claude 2.1 is live with a new increased 200k context window. Matt Schumer writes, If you're an OpenAI developer worried about what will happen to your apps if OpenAI goes down, this is a fantastic alternative. And so that's where we are, waiting to see what's next.

17:02Kara Swisher summed up, Think you just have to see some move tonight or by tomorrow by OpenAI board, who better have some good lawyers and better explanations. Right now, we've heard from neither, and so I guess we just have to wait. That's going to do it for today's AI Breakdown. Appreciate you listening as always. And until next time, peace.