OpenAI's New System for Determining How Close AGI Is

0:00Today on the AI Daily Brief, we discuss the OpenAI's new system for determining how close to AGI we actually are. Before that, on the headlines, we talk about Intuit's firing 1 ,800 people and then rehiring 1 ,800 other people to better compete in AI. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes. Welcome back to the AI Daily Brief Headlines Edition, all the AI headline news you need in around five minutes. We kick off today with an interesting story around how one big company is figuring out the AI transformation and transition.

0:37We're very keyed up to see any time layoffs happen because of AI. However, Intuit's story is a little bit more complicated. The headline is that the company is laying off 1 ,800 people, obviously a major impact. It's around 10 % of their workforce. And the culprit is indeed artificial intelligence. Said Intuit CEO Sasan Ghadarzi in a letter,

1:15So what then is Intuit doing? Well, they write, I am sharing some very difficult decisions my leadership team and I have made to ensure we are able to increase investment in these key growth areas. Today, we will be communicating to approximately 1 ,800 employees, which is 10 % of our workforce, that they will be leaving into it. Sasan continues, We do not do layoffs to cut costs, and that remains true in this case. The changes we are making today enable us to allocate additional investments to our most critical areas to support our customers and drive growth. This includes reinvesting in the necessary skills and capabilities to support these areas, and as such, we will hire approximately 1 ,800 new people, primarily in engineering, product, and customer-facing roles such as sales, customer success, and marketing.

1:57At least that's how they're framing it. Now, digging in, it seems like it's about a little bit more than AI. Intuit is closing two offices, one in Edmonton, Canada, and one in Boise, Idaho, affecting 250 employees. Another 1 ,050 are getting laid off because they're, quote, not meeting expectations. They've eliminated 300 roles to support streamlining operations and shifting resources towards AI. And they're consolidating 80 tech roles to quote sites where we are strategically growing our technology teams and capabilities. The blog post is interesting for how much detail it goes into around their strategy.

2:28Ultimately, though, what people are taking away is that in the era of AI, we are going to see some big shifts in how companies structure how they pursue their goals. So it's likely that we see a lot more like this. Speaking of big old companies and transitions, one story that we've been tracking is a potential deal between Elon Musk's XAI and Oracle, which, if completed, would give XAI access to more computing power. Writes the information, Larry Ellison may be a friend and backer of Elon Musk, but even the Oracle executive chair isn't immune for Musk's impatience. According to the information, the companies have ended talks to expand an existing relationship, under which XAI rents NVIDIA chips from Oracle.

3:04Instead, XAI is going the solo approach, buying more chips to build a data center of its own. The deal had been worth a reported$10 billion. But according to the information again, quote, the talk stalled over issues including Musk's demand that the supercomputer be built faster than Oracle thought possible, and Oracle's concern that XAI's preferred location did not have access to an adequate power supply. Now, XAI is moving ahead with a plan to build its own data center in Memphis, Tennessee, with NVIDIA chips that are supplied by Dell and Supermicro. Said Musk in a post on X, Oracle is a great company, but when our fate depends on being the fastest by far, we must have our own hands on the steering wheel rather than be a backseat driver.

3:42Over in the policy world, even more concerns about that Microsoft G42 deal. You'll remember that when it was announced, and by it I'm referring to a$1.5 billion minority investment from Microsoft into G42, it was announced in part by the Department of Commerce. They had clearly been involved in negotiating the deal. And it read to me, frankly, like the government using Microsoft as a way to get a peek into G42's books. Now, the concern around G42 has always been the fact that they, more than any other company, represent the Middle Eastern Gulf region's attempt to straddle the gap between the U.S.

4:13and China. But now, even though the government was involved in that deal, numerous lawmakers are raising additional concerns. Representative Mike McCall, the chair of the House Foreign Affairs Committee, and John Moulinard, the leader of the Select Committee on China, have requested that the Biden administration makes an intelligent assessment of Microsoft's investment in a letter that was dated this Wednesday and addressed to White House National Security Advisor Jake Sullivan. Said the letter, we remain deeply concerned by attempts to move quickly to advance a partnership that involves the unprecedented transfer of highly sensitive U.S.

4:43origin technology without congressional consultation or clearly defined regulations in place. For those who don't spend a ton of time paying attention to the battles between the administrative state and Congress, this is a growing theme across numerous issues right now. Elected officials in the House and Senate are increasingly trying to claw back powers that have been assumed by appointed officials, and indeed, many of the cases that deal with these issues are being picked up by the Supreme Court. TLDR of this all being, then, that not only is AI ending up a geopolitical issue with the U.S. and China, it's also increasingly becoming an issue as relates to broader power struggles happening right here at home as well.

5:19Lastly today, a cool one to end our week. BBC writes that the sperm whale's phonetic alphabet has been revealed by artificial intelligence. Researchers studying sperm whale communication say they've uncovered sophisticated structures similar to those found in human language. The BBC writes, In the inky depths of the Midnight Zone, an ocean giant bears the scars of the giant squid she stalks. She searches the darkness, her echolocation pulsing through the water column. Then she buzzes, a burst of rapid clicks, before she goes in for the kill. but exactly how sperm whales catch squid like many other areas of their lives remains a mystery.

5:51Now the study that would lead to some of these insights actually started all the way back in 2005. That's when the Dominica Sperm Whale Project was formed to study the social and focal behavior of around 400 sperm whales that live in the Eastern Caribbean. What's happened is that, taking advantage of almost 9 ,000 recordings, AI has helped researchers identify 156 distinct rhythmic sequences of clicks called codas. Again from the BBC, the sperm whales were found to vary the overall speed or the tempo of the codas as well as to speed up and slow down during the delivery of a coda. Sometimes they add an extra click at the end of a coda, and these subtle variations suggest sperm whale vocalizations could carry a much richer amount of information than previously thought.

6:30We are still at the very beginning of the way that AI is going to help transform research and scientific discovery, and already it is so cool. That, however, will do it for today's headlines edition. Next up, the main episode. Today's episode is brought to you by Superintelligent, the platform for fun, fast AI learning. Super has a ton of new things going on. We recently announced our partnership with Spotify, through which users of that app can now access Superintelligent content directly from their mobile apps. We've also just launched the AI learning feed. In addition to seeing the tutorials that we're dropping, there are polls, news items with related lessons, and a chance for people to show off the projects and use cases that are making AI come alive for them.

7:11We've also just kicked off the Super Summer Challenge, where each week we'll share a new challenge that you can use to discover new AI tools and use cases. Go to bsuper.ai and use code SUPERFUN for 50 % off your first two months. That's bsuper.ai. Today's episode is brought to you by Venice. Venice is a private, uncensored generative AI app. It accesses open source models to enable text, image, and code generation without the fear of being spied on or having your data exploited. Discuss anything with Venice without concerns about it being monitored, sold, or given to advertisers and governments.

7:43Venice is different because your conversations and creations are kept securely within your own browser, never stored or accessible by Venice. Unlike other AI apps, Venice won't tell you what's okay to say or not. Venice won't patronize you. It simply provides direct access to machine intelligence. No topics are off-limits, no ideas are taboo. With Venice, you're in control of the AI, as you should be. Pro subscriptions are available for$49 a year or$8 per month. Try it for free without an account at venice.ai. Welcome back to the AI Daily Brief. Today, our main episode focuses on two stories coming out of OpenAI.

8:17They are very different, but both are really big and fascinating in their own ways. We kick off with a new report, a scoop from Bloomberg, really, that OpenAI has developed a new system to explain where we are in progressing towards AI that achieves artificial general intelligence or AGI. It is, of course, the expressed mandate and purpose of OpenAI to produce safe and beneficial AGI for all. But this is new territory, and so benchmarking where we are against it theoretically seems like a pretty useful thing. Ever since the crisis last fall with Sam Altman being fired and then rehired, and more recently with Ilya leaving the company alongside basically the entire super alignment team, it's been kind of a wait and see on how OpenAI would talk again about its approach to long-term AI alignment and AI safety.

9:05This new leveling system certainly isn't a comprehensive approach to that, but does give a little bit more insight into how the company sees risks at various stages. The new system has five levels and was introduced at an all-hands meeting on Tuesday of this week. It was confirmed by an OpenAI spokesperson. So what are these five levels? According to OpenAI, the stages of artificial intelligence start where we are now, at level one with chatbots. These are AIs that can use conversational language. Of course, we are all living in the post-Chat GPT world, but fundamentally what OpenAI is saying with these levels is that for almost two years now, this is the level that we've been at, level one.

9:44Presumably then, level one does not represent some big risks. There isn't really anyone out there saying that this sort of GPT-4.0, Claude 3.5 type of capacity is a real risk. But what comes next? The level two stage on OpenAI's system is called Reasoners with human-level problem solving. According to Bloomberg, at the meeting where they introduced this leveling system, execs at the company told employees that they believed that while we were on the first level, we were on the cusp of reaching the second. Adding a little bit more color, Bloomberg writes, this refers to systems that can do basic problem-solving tasks as well as a human with a doctorate-level education who doesn't have access to any tools.

10:22To point out how this version would differ, Bloomberg writes, Company leadership gave a demonstration of a research project involving its GPT-4 AI model that OpenAI thinks shows some skills that rise to human-like reasoning. This part was not confirmed by the OpenAI spokesperson, who simply said that they are, quote, always testing new capabilities internally. Now, you may note that Level 2 is not the thing that everyone has been excited about for more than a year, which is agents. That comes at Level 3. The Level 3 stage is agents or systems that can take actions. At level four, we have innovators, AI that can aid in invention.

10:56And at level five, organizations, AI that can do the work of an entire organization. A couple more caveats. This is not necessarily a final determination. This was created by a set of OpenAI executives and is meant to be a work in progress. Second, it doesn't appear at least that we've seen any sort of discussion around how long OpenAI believes it will take to get to each of these different stages. I will note that I did see on Twitter today someone who claimed that they had sources in their DMs saying that they had actually interacted with GPT-5, although that tweet came down later, so make of that what you will.

11:28If you head on over to Metaculous, which is an interesting place to get a crowdsource sense of when people who are pretty well informed think different milestones will be hit within AI, the current betting among 652 forecasters around when OpenAI will announce GPT-5 is January 13th of next year, 2025. I will also say that while we've only started to see conversation around this system hitting x slash twitter people aren't necessarily totally convinced or at least they don't feel like they have enough information to really understand mario canistra writes strange list it seems to me that as soon as you get to level two you also get three four and five unless there is a weird definition of human level humans can do all these things so human level ai should be able to do all those things others like matt garcia points out that level five organizations kind of sounds like advanced super intelligence Pseudonym writes, this would be a better way to structure an at-home agent flow.

12:18And Steph writes, once you reach level two, level three should be a piece of cake. I'm probably wrong, but just by making AI have access to APIs, it should become an agent if it can already plan things. I will be interested to see whether anyone else picks up this discussion over the next couple weeks, or whether this will mostly remain on the drawing board. Second OpenAI story, one that has much more to do with the state of US policy in AI, Microsoft and Apple are both getting rid of their OpenAI board observer roles. So first, a little bit of background. Microsoft has obviously had a long-standing relationship with OpenAI, but it wasn't until after the debacle last November that it took a formal board observer role.

12:55More recently, really just a week ago, we learned that Apple would be getting an equivalent board observer role as part of its deal to put ChatGPT onto the iPhone. Now, however, it appears that antitrust concerns have made those big tech companies rethink that relationship. The reality is that we have very little information around what drove this. Presumably, it has to do with antitrust scrutiny, which has been on the rise in general. Writes Martin Peers at the information, Given that Microsoft's board observer role seems to have intensified regulatory scrutiny of its OpenAI relationship, both in Europe and in the US, it's possible Microsoft and Apple decided the positions weren't worth the hassle.

13:30OpenAI can update both companies on whatever is necessary in private meetings as it indicated it would do going forward. So here, basically, Martin is arguing that a board observer role is exactly that. It's an observer role. Doesn't have any power to actually shape or influence things or be a part of any sort of voting. And so to the extent that that role is just about visibility and insight, but it increases regulatory pressure, well, just figure out some way that isn't enshrined as a board observer to do the same sort of information sharing. Piers does also note that given the European Commission's antitrust leader had said that her team had concluded that Microsoft hadn't acquired control of open AI, after getting this board observer role, it quote, implies that if regulators triggered this shakeup, it was either the FTC in the US or the Competition and Markets Authority in the UK.

14:15While the regulatory discourse around AI has certainly slowed down in the US this year, the anti-drust conversation has done nothing but heat up. For now though, that is going to do it for this quick episode today. Appreciate you listening as always. And until next time, peace.

14:36Thank you.