The AI Daily Brief: Artificial Intelligence News and Analysis - Altman Says AGI Coming Faster Than We Think

Episode Date: December 6, 2024

Sam Altman shares bold predictions about AGI during the DealBook Summit, claiming it's coming sooner than many anticipate. This video covers Altman's remarks, including OpenAI's progress on AGI, updated user stats for ChatGPT, and insights into OpenAI's evolving relationship with Microsoft. The discussion also covers OpenAI's defense partnership and the highly anticipated "12 Days of Shipmas" announcement series. Brought to you by: Vanta - Simplify compliance - ⁠⁠⁠⁠⁠⁠⁠https://vanta.com/nlw The AI Daily Brief helps you understand the most important news and discussions in AI. Subscribe to the podcast version of The AI Daily Brief wherever you listen: https://pod.link/1680633614 Subscribe to the newsletter: https://aidailybrief.beehiiv.com/ Join our Discord: https://bit.ly/aibreakdown

Today on the AI Daily Brief, AGI might be her sooner than we think, and 300 million people are now using chatchapit every week. Before then in the headlines, Google CEO seems to acknowledge the AI slowdown. 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 at our show notes. Welcome back to the AI Daily Brief Headlines edition, all the daily AI news you need in around five minutes. One of the things that we have been keeping an eye on recently is whether the idea that there is, a slowdown or a plateau in AI capabilities actually has legs. Well, in a recent interview, Google CEO Sundar Pichai seems to give some credence to that. Talking about the company's

AI research efforts, Pichai said, the low-hanging fruit is gone. The hill is steeper. The information continues. Pichai said models would continue to get better at completing a sequence of actions, which would help with automating software tasks, but quote-unquote deeper breakthroughs would be needed for more dramatic improvements. This is certainly the theme that we've been hearing from lots of different folks. And of course, we've covered a number of the different approaches to scaling that organizations are now experimenting with. I was giving context earlier today, though, for why none of this particularly freaks me out or makes me think that anyone can really rest on their laurels. Dan Shipper from Every tweeted this morning, here's the most compelling benchmark of AI progress,

a task that took 60 minutes a year ago now takes 60 seconds. In January, 24, Jeffrey Litt and I spent an hour coaxing ChatGBT and Replit to build a live app on my podcast. 12 months later, Steve Rouse and I built the same app with one prompt in less than a minute. Now, keep in mind, building an app live with chat GPT and Replit in an hour was a very impressive thing. It still is a very impressive thing. So the fact that a year later, less than a year later, really, that same app is getting built in a single prompt shows just how much we have to get out of the current state of the art. Staying on the Google theme for a minute, one of that company's big breakout AI products for this year is, of course, Notebook LM. Really interesting collaboration on that front, Spotify and Notebook

LM are embedding a personalized AI podcast as part of Spotify's annual RAPT feature. RAPT is Spotify's summary of your personal listening habits, and this year they included a podcast version of the recap built of course with Google's notebook LM. Very smart partnership for not only providing a cool alternative interface for RAPT, which is an extremely popular feature, but also for exposing more people to notebook LM. Speaking of notebook LM, some of the people involved with that feature are using the momentum to do something different. Three core members of the team have left the company to strike it out on the

their own. Team leader Riza Martin, along with designer Jason Spielman and engineer Stephen Hughes, have resigned to found a new stealth startup. In a LinkedIn post, Martin said that each of the three, quote, have always wanted to be part of a startup and we couldn't shake the feeling that there's a massive opportunity to build something transformative in the space. At this stage, we just have an empty website at we're building.a.i. But Martin did hint at another app player project stating, as the frontier models and their capabilities continue to grow, thoughtful products are required to make the benefits of this technology accessible, useful, and obvious to everyday people. So our team is going to be focused on building a user-first AI product.

She said the company is in its earliest stages and has not announced any funding yet, although you have to think, given how successful notebook has been, that funding will not be an issue for this particular startup. One more from Google Land, really dominating this headline section today. Google DeepMind has unveiled their world model called Genie 2. In an announcement thread, team lead Jack Parker Holder said, Introducing Genie 2, our most capable large-scale foundation world model, which can generate a diverse array of consistent worlds playable for up to a minute.

We believe Genie 2 could unlock the next wave of capabilities for embodied agents. Basically, the tech promises to deliver consistent environments with simulated dynamics, playable with a mouse and keyboard. Parker Holder also showcased the idea of embodied agents or AI-controlled avatars that can navigate the generated worlds. He said that the combination of technologies could be used as a, quote, path to unlimited environments for training and evaluating our embodied agents. Now, this is a huge theme that we're seeing more and more.

Earlier this week, we saw what Fei-Fei Lee's World Labs has been working on with their own world model that can generate a 3D environment based on a single image. DeepMind's model works in the same way, but seems to have focused much more on video game-style features. In a blog post, they wrote, Genie 2 responds intelligently to actions taken by pressing keys on a keyboard, identifying the character and moving it correctly. For example, our model can figure out that arrow keys should move a robot and not trees or clouds. The Genie 1 model had been an extremely narrow experiment, focused only on generating 2D worlds. DeepMind said this model was a significantly forward in generality, adding, Genie 2 is a world model, meaning it can simulate virtual worlds, including the consequences of taking

any action. It was trained on the large-scale video dataset, and, like other generative models, demonstrates various emerging capabilities at scale, such as object interactions, complex character animation, physics, and the ability to model and thus predict the behavior of agents. I think for the Long Reads episode this weekend, we might get a little bit deeper into these world models, so keep an eye out for that. Lastly today, meta is getting in on the nuclear.

game. Joining Amazon, Microsoft, and Google, Meta announced a request for proposals to suit their nuclear ambitions. The company is seeking nuclear energy developers who can deploy between one and four gigawatts of capacity in the U.S. The new generation would be used to power Meta's data centers, but also had a strong green energy angle. Meta wrote, supporting the development of clean energy must continue to be a priority as electric grids expand to accommodate growing energy needs. At meta, we believe nuclear energy will play a pivotal role in the transition to cleaner, more reliable, and diversified electric grid. They are looking to a establish the new supply by the early 2030s. The timeline is a little tight, with construction of the last

new nuclear facility to be built running seven years over. Then again, meta is pretty flexible on the details. The new power plant could be built anywhere, rather than needing to hook up to an existing data center. And with a string of regulatory hiccups delaying big tech's nuclear ambitions, this proposal could open up a new avenue. As well as being a casting call for nuclear developers, the proposal could entice states to offer a regulatory fast track to attract new infrastructure. Separately, META announced plans to build a supercluster in Louisiana. In a post on threads, the company wrote, Today, in partnership with Louisiana Economic Development,

we announced our plans for the newest and largest data center in Richland Parish, Louisiana, which will play a vital role in accelerating our AI progress and support training future open source LLMs. The state's governor announced the project for the budget will be $10 billion, with construction to commence this month and continue through to 2030. But that is going to do it for today's AI Daily Brief Headlines edition. Next up, the main episode. Today's episode is brought to you by Plum.

Want to use AI to automate your work but don't know where to start? Plum lets you create AI workflows by simply describing what you want. No coding or API keys required. Imagine typing out AI, analyze my Zoom meetings and send me your insights in Notion and watching it come to life before your eyes. Whether you're an operations leader, marketer, or even a non-technical founder, Plum gives you the power of AI without the technical hassle. Get instant access to top models like GPT40, Claude Sonnet 3.5, assembly AI, and many more.

Don't let technology hold you back. Check out Use Plum, that's Plum with a B, for early access to the future of workflow automation. Today's episode is brought to you by Vanta. Whether you're starting or scaling your company's security program, demonstrating top-notch security practices, and establishing trust is more important than ever. Venta automates compliance for ISO-2701, SOC2, GDPR, and leading AI frameworks like ISO-402, and NIST AI Risk Management framework, saving you time and money while helping you build customer trust. Plus, you can streamline security reviews by automating questionnaires and demonstrating your security posture with a customer-facing trust center all powered by Vanta AI. Over 8,000 global companies like Langchain, Lila AI, and factory AI use Vanta to demonstrate AI trust and prove security in real time. Learn more at vanta.com slash NLW. That's Vanta.com slash NLW. Today's episode is brought to you by Super Intelligent. Every single business workflow and function is being remade and reimagined with artificial intelligence.

There is a huge challenge, however, of going from the potential of AI to actually capturing that value. And that gap is what Super Intelligence is dedicated to filling. Super Intelligence accelerates AI adoption and engagement to help teams actually use AI to increase productivity and drive business value. An interactive AI use case registry gives your company full visibility into how people are using artificial intelligence right now. Pair that with capabilities building content in the form of tutorials, learning paths, and a use case library. And Super And Super Intelligent helps people inside your company show how they're getting value out of AI, while providing resources for people to put that inspiration into action.

The next three teams that sign up with 100 or more seats are going to get free embedded consulting. That's a process by which our Super Intelligent team sits with your organization, figures out the specific use cases that matter most to you, and helps actually ensure support for adoption of those use cases to drive real value. Go to BSUPER.AI to learn more about this AI enablement network. And now back to the show. Welcome back to the AI Daily Brief. Well, I fear that we may be having quite a bit of open AI talk coming up.

Yesterday, Sam Altman tweeted, starting tomorrow, we are doing 12 days of open AI. Each weekday, we will have a live stream with a launcher demo, some big ones, and some stocking stuffers. We've got some great stuff to share. Hope you enjoy. Merry Christmas. I'll come back in a minute to what people think might be coming from that. But where I actually want to start is with a really interesting conversation. between Sam Altman and Andrew Ross Sorkin at the New York Times Deal Book Summit yesterday. There were a bunch of really fascinating parts of the conversation, including some new numbers

for chat GPT revealed, as well as maybe the most notable part, which is how Altman thinks about the coming of AGI. The TLDR is that Altman thinks that AGI is coming a lot faster than we think, but maybe less impactful than we think. He said, my guess is we will hit AGI sooner than most people in the world think, and it matters much less. AGI can get built. The world mostly goes on in mostly the same way. The economy grows faster, but then there is a long continuation from what we call AGI to what we call superintelligence.

Last month, Altman said he believes his company will achieve AGI next year, and these comments suggest he thinks Open AI is still on course. The big question, though, is how exactly does Open AI and Altman define AGI? An early version of the company charter said AGI would be able to, quote, automate the great majority of intellectual labor. Now it seems that the definition is a little fuzzier and somewhat less. One of the reasons this matters is that OpenAI can end their deal with Microsoft once their board declares that AGI has been achieved. The Verge recently reported that the goal seems to be to combine the forthcoming Orion or

GPT5 model with the O1 reasoning model to create a more powerful model that can be labeled AGI. Adding more credence to the idea that Altman's definition of AGI is now more modest, he added, I expect the economic disruption to take a little longer than people think because there's a lot of inertia in society. So in the first couple of years, maybe not that much changes. and then maybe a lot changes in the economy. It may be that the civilization changing AI

which can end the need for human work is what Albin has been referring to now as superintelligence. Alman wrote back in September that superintelligence could be achieved within a few thousand days, a much less certain timeline. And notably, while Alman was optimistic that they're close, he did not give the impression

that Open AI has AGI waiting in the wings almost ready to deploy. He said, there's a ton of hard work, a ton of research and engineering still to do, but I think it's possible. The other big topic from the interview was the prospect of political retribution from Elon Musk. So far, Musk has been fighting OpenAI in the courts, but there's a growing concern that he will use his newly found political influence to stifle

Altman's plans. Discussing the prospect that Musk could send the force of the government against Open AI, Albin said, I believe pretty strongly that Elon will do the right thing, and that it would be profoundly un-American to use political power to the degree that Elon would hurt competitors and advantage his own businesses. I don't think people will tolerate that, and I don't think Elon would do it. Alman also remarked that he was, quote, tremendously sad about the simmering tension between the pair of one-time co-founders, commenting, I grew up with Elon as like a mega-hero. I thought what Elon was doing was absolutely incredible for the world at a time when most of the world was not thinking very ambitiously. He pushed a lot of people, me included, to think, to think, to think, he's not much more ambitiously. And I have different feelings about him now, but I'm glad he exists. Reflecting on what is driving Elon to go after Open AI in this way, Alman seems to think it's purely about business rivalry, adding, he really cares about being the guy. He's a competitor, and we're doing well. There was also a lengthy discussion about the conversion of OpenAI into a for-profit company.

Altman confirmed that the decision was made in part because Musk caught off funding, but didn't elaborate on the events that led to the situation. He denied reports that OpenAI had sought to block investors from also investing in rival companies. Altman also discussed his lack of equity in the company stating, I wish I had taken some. Given how big all of this is, it almost passed without notice that Chad GPT usage has exploded. Allman claimed that the platform now sees 300 million weekly active users. Basically, the equivalent of a major portion of the entire United States is using chat chit

every week. They're sending a billion messages daily, and 1.3 million developers in the U.S. alone are building on the platform. In the context of the abnormally fast scaling, Altman noted that the company requires far more compute than they had projected. And on the rumors that the relationship with Microsoft had become frosty, he admitted to some tension but said that the two companies were still pretty aligned on their incentives and are not in the middle of disentangling.

One other interesting story around OpenAI that isn't about their shipmiss, which is where we'll end this episode, the company assigned a partnership with Anderil industries to build anti-dron defense systems. This is OpenAI's first time working with a commercial weapons manufacturer

and comes a month after they first signal their newfound willingness to work on military tech. The company still prohibits the use of its technology on offensive weapons, but has become increasingly comfortable contracting with the Defense Department on cybersecurity and administrative work. The current crop of Silicon

Valley startups is undeniably more willing to view tech as an integral part of national defense and something they should be taking an active role in. Last month, Anthropics signed a deal with Palantir to provide their models for classified document analysis. Anderil's anti-dron systems are designed to track unmanned aircraft and then shut them down using electronic jamming, response drones or other means. Anderle said that OpenAI's tech could improve accuracy and reduce the time to detection and response, putting fewer people in harm's way. Pretty interesting one that I think says a lot about the shifting mood of Silicon Valley. Lastly, let's talk about 12 days of Shipmiss.

An announcement tweet promised 12 daily live streams and a bunch of new things big and small. The Verge reported that the stocking stuffers will include an assortment of new features, products, and demos. And sources say that the big reveals will be a new reasoning model and the long-awaited release of the text-to-video model Sora. It seems that some shipmiss cheer will also be brought to the company's existing products. AI news aggregator B. Tibor found code enabling a new Christmas edition voice for ChatGBT, complete with a snowflake symbol to activate voice mode. DeBoer is referring to it as Santa GPT, so perhaps you'll be able to chat with OpenAI's version of jolly old St. Nick.

For what it's worth, this is something that I have already done with my kids asking one of ChatGPT's existing voices to imitate Santa to announce as we turned on our holiday lights. Overall, the team at OpenAI seems to be very excited about this. Chief Product Officer Kevin Weill tweeted this is going to be wild and crazy and a lot of fun. Shipmiss is coming. Investor Vinod Kosla wrote, excited for OpenAI's 12 days of shipmiss. The AI world will again feel different by the new year.

We of course got a bunch of predictions. Data scientist Diego wrote, day one, launch of SORA. Day two, introduction of a Santa-inspired voice for ChatGBT.D.T. Day three, enhancement of chat UPT's advanced voice mode with vision. And then he went on to make a bunch of predictions as well. One small one, which would be big, would be a new Dali model. To the extent that there is any skepticism, it's that some think that if the announcements aren't really big, this is all going to feel like bit of a wet blanket. Lissano Gabe writes, I think I'm speaking for most AI nerds when I say, if OpenAI doesn't release GPT4.5 or demo GPT5, this 12-day event will be a huge waste of time and could really screw Open AI's public perception. We already know that Sora-O-1 price cuts in AVM are

coming because you promised us months ago. Give us a Christmas surprise or pack your stuff. If you make all this hype and don't deliver, I think as an investor, I would rather bet my chips on Claude, Gemini, Grock, Deep Seek, and Quinn. For my part, I do not think we're going to see GBT 4.5 or GPT5. Alman has been pretty clear that that is not coming this year. However, I think a lot of people will be pretty excited if SORA actually comes to bear. And as for the rest of it, well, we'll just have to wait and see. For now, though, that is going to do it for today's AI Daily Brief.

Until next time, peace.