The AI Daily Brief: Artificial Intelligence News and Analysis - OpenAI Raising at $90B and Trolls Everyone, Saying AGI Has Been Achieved

Episode Date: September 27, 2023

OpenAI man. The company is reportedly offering employees a chance to sell shares between an $80b-$90b valuation; discussing an AI hardware device with iPod designer Jony Ive; and scaring the hell out of people with a troll that "AGI has been achieved internally." Before that on the Brief: Apple is the most AI acquisitive Silicon Valley giant, and the CIA is working on a chatbot for intelligence agencies. ABOUT THE AI BREAKDOWN The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, we're looking at news that OpenAI might be raising at a $90 billion valuation. Before that on the brief, the CIA is making a chat GPT style service, and Apple is quietly the biggest AI acquirer in Silicon Valley. 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 Discord, our YouTube channel, and our newsletter. Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five minutes. We kick off today with an interesting story around Apple. Now, Apple is, of course, the big open question mark when it comes to big tech and generative AI. The company has clearly

not been as willing as its peers to throw itself headlong into at least the buzziness of the gen AI space, but there are a lot of indications that they are doing more than perhaps they're letting on. Recently, for example, the information reported that Apple is spending millions of dollars a day now training AI models, even if they're not exactly sure what they're going to do with them. We've also gotten glimpses of how Apple might be thinking about implementing AI with recent product launches of the iPhone and the latest Apple Watch. The Apple Watch in particular uses a new chip with what they call a neural engine for better inference on device. Now, they're using that to make gesture-based controls viable for operating the Apple Watch with a single hand. But the broader applications are, of course, AI models that can actually run on device without touching the cloud.

Many have speculated that that will be the differentiation that Apple tries to seek, given their existing stance on privacy and user data. Now, yesterday we got a report from Pitchbook, summed up by Quartz, which showed another indicator of how active Apple actually is in the AI and machine learning space, even if it's not being flashy about it. Since 2017, Apple has bought by a distinct amount more AI and ML companies than its peers like Microsoft and Google. In fact, since 2017, Apple has bought 21 AI startups. Next on the list is consulting Giant Accenture with 19.

But after that, it's Microsoft at 12, meta at 11. So in each case, just a little over half of what Apple has acquired. and Google's alphabet is all the way down the list at just eight. Now, for now, Apple remains intransigent about talking about its AI strategy. Indeed, when asked on a conference call with investors and analysts in August why Apple was being so tight-lipped about its AI investments, CEO Tim Cook said, we tend to announce things as they come to market, and that's our MO.

And I'd like to stick to that. Now, speaking of people bringing AI tools to market, it will surprise perhaps no one to find that the USCIA has built its own chatchip-t-style AI tool with a specific eye to continue combating the rise of China. Basically, the CIA has something called the Open Source Enterprise Division, and they're planning to provide all 18 U.S. intelligence agencies with a chat-GPT-style tool that can sift through huge, huge volumes of public information, and answer questions in a conversational way. Randy Nixon, who leads that division, said, we've gone from newspapers and radio,

to newspapers and television, to newspapers and cable television, to basic internet, to big data, just keeps going. We have to find needles in the needle field. Now, obviously, anyone who's used chat GPT can see why this would be an appealing use case. What ultimately is an LLM like chat GPT good at? Well, one, it's good at aggregating an incredible volume of information, and two, it's good at letting people use natural language to recall and better access more quickly that information. It's not hard to imagine the intelligence applications there. Now, some concerned observers have noted, however, that the CIA has not revealed what model will be used in training the tool, exactly what information will be available to it, or how it will protect

that information from accidentally getting public. Darknet Diaries podcaster Jack Reisiter sums up the concern that I've seen on Twitter by saying, OK, I'm confused. Define available data. Does the CIA need a search warrant to search Twitter posts? Or how about my SMSes? Are my tax records in this mess? At Just Matthew responds, assume it means everything.

Now, staying on this theme of the military use or the intelligence use case of AI, the news outlet Niki is reporting that the U.S. is preparing to pursue a UN resolution on international norms for how AI is used in weapons systems. Now, in terms of what will be included, we don't exactly have a lot of information yet. However, observers expect whatever finds its way into the resolution to be based on a document published in February by the State Department called Political Declaration on Responsible Military Use of Artificial Intelligence and Autonomy. That document includes a number of things such as using military AI in line with international humanitarian law, suggesting each country issue its own principles for using AI systems, calling for AI to be disabled if it starts to operate in

unintended ways, and urging human involvement for any and all actions relating basically to anything with nuclear weapons. Now, one State Department diplomats said that the U.S. had actually been discussing the declaration not just with its traditional allies, but with a much wider range of companies, and said, quote, we've gotten a lot of positives. We've had comments on it and some ideas for changes to text, but we haven't had a lot of pushback. That same official Bonnie Jenkins said that when it came to engagement with Beijing in this area, that quote, this is an area that we would like to work with China. Moving to a different dimension of artificial intelligence in the real world, yesterday we reported that there had been a tentative agreement reached between the Hollywood studios and the writers,

and subsequent to that recording, we started to get a few more details about what the actual AI aspect of that deal would include. While we still don't have the exact language, it sounds like the writers walked away with a guarantee that they could receive credit and compensation for work they do on scripts, even if studios partially rely on AI tools, but that at the same time, Hollywood Studios retained the right to train AI models based on writers' work. One person who didn't like the deal was IAC chair Barry Diller. CNBC writes, Barry Diller rips WGA deal with studios says fair use needs to be refined, saying it doesn't do enough to address the AI threat. Appearing on CNBC's Squackbox, Diller said, fair use needs to be redefined because what they have done is sucked up

everything and that violates the basis of copyright law. All we want to do is establish that there is no such thing as fair use for AI, which gives us standing. Now, when it came to this deal, he said, they spent months trying to craft words to protect writers from AI, and they ended up with a paragraph that protected nothing from no one. So there will be a lot to continue to watch in this situation as it evolves. In the world of Enterprise AI, Enterprise Giant SAP has launched their own AI assistant called Jewel. The play here is what we've talked about innumerable times on this show, a company that has extensive enterprise relationships and customers already, plugging AI into the existing apps and services that those customers already use. One of the big questions for me remains whether

when it comes to these enterprise tools, there will become a dominant standard, or whether the chatbots and the LLMs that underpin them will become eventually totally commodified, and companies will just use whatever service providers they already used and had existing data relationships with. As we start to round out this show, one other update from a story earlier this week, we talked a bunch about Spotify's AI tools for podcast translation. But what we didn't mention is that in a recent BBC interview, Spotify's CEO Daniel Eck had said that Spotify had changed its opinion a little bit and was trying to take a more nuanced approach that stopped short of completely banning AI music. Eck basically said that there are valid uses of AI in music that should be allowed, but that

where to draw those lines was going to be, quote, tricky. Eck said we have a very large team that is working on exactly these types of issues. It seems clear to me that when it comes to music creation, there has to be some amount of give here, and I think that the most likely scenario is basically sanctioned whitelisted assets in the form of specific music that people can train models on that gives the bedroom creators the ability to do what they were going to do anyway, but actually cutting the artists and the labels in on the deal. On the AI copyright front, it looks like we are about to get a jury trial around the Thompson Reuters' AI dispute. writes Reuters, quote,

A jury must decide the outcome of a lawsuit by Information Services Company Thompson Reuters accusing Ross intelligence of unlawfully copying content from its legal research plan. platform Westlaw to train a competing artificial intelligence-based platform. Now, of course, there are numerous other AI model training-related lawsuits out there, but this could be the first one to come to a trial. Anyways, guys, another busy day in the world of AI. There were so many things that I didn't even get to in this. Stick around, because coming up next, we have a look at all of the news around OpenAI,

including a big potential fundraise, a collaboration with the creator of the iPod, and much, much more. Hey, guys, one more quick thing before we get into the main episode. you subscribe to the newsletter, you've seen this, and you might have heard it on an earlier episode, but right now I am getting information from you guys, the listeners, about what you are looking for in terms of AI educational resources. A bunch of you have filled out the survey already, and it's so helpful. But if you would take the about one minute and go to bit.ly slash AI breakdown survey, I would love to know what type of online courses you might need, what you're trying to

learn more about, whether you'd be interested in a community of learners. I'm getting really close to making some decisions about what we're going to do next, and I really want all of your input. Again, it'll take about one minute, and you can find it at bit.ly slash AI breakdown survey. Thanks so much, and now on with the show. Welcome back to the AI breakdown. So as you could probably guess from that introduction, today's topic or set of topics, really, is all about open AI. There was a bunch of news over the last 24 hours, and we're going to go through it piece

by piece, starting with the big headline that the company is exploring a new capital raise at a significantly marked up valuation. So by way of background, OpenAI is of course 49% owned by Microsoft and has raised $11 billion so far. The big $10 billion Microsoft investment was done at a post-money valuation of around $30 billion, which is how they own about half of the company. So raising at between $80 or $90 billion would represent some serious paper gains, if nothing else, for Microsoft. Now, it doesn't appear to me, based on the little information that we have, that this deal is really about Open AI trying to add to their war chest. Instead, it looks to me like a combination of, one, wanting to give some early employees a little bit of liquidity on their shares,

and two, giving a set of new investors who aren't on the cap table yet a chance to get involved. The Wall Street Journal writes, the deal is expected to allow employees to sell their existing shares as opposed to the company issuing new ones to raise additional capital. Now, why would this make sense? well, of course, AI is just about the hottest labor market that exists. There is a constant back-and-forth flow of talent between the big players, incredibly inflated salaries, and just in general, a huge competition for the best talent. OpenAI opening up this additional investment opportunity potentially gives them the ability to incentivize employees by showing that they're committed to helping them get liquidity on their equity earlier than they might otherwise have.

That could be a strong inducement, for example, for people to stick around to continue vesting. Now, if they do achieve that $80 or $90 billion valuation, the WSJ points out that that would make them one of the most valuable startups in the world behind only basically SpaceX and TikTok, or rather TikTok owner bite dance. In terms of total amount, the raise would be on the order of hundreds of millions of dollars, not billions. And the one new investor that has been mentioned frequently in the press is, of course, SoftBank. Now, one open question is whether SoftBank would come in just through this employee tender offer, or whether there would be an entirely new round priced at the minimum price of this employee share, say, but representing a bigger raise overall. Now, we've heard recently a lot about SoftBank and its desire to go even deeper in the AI space. The investment giant made a ton of money in the RMIPO and reportedly wants to redeploy that capital into the artificial intelligence space. On September 16, Reuters wrote a piece called SoftBank seeks OpenAI tie up as Sun plans deals

free after RMIPO. That article suggested that Masayoshi Sun, the head of SoftBank, of course, speaks, quote, almost every day to OpenAI CEO, Sam Altman. Now, in the wider community, this has raised questions of whether we are now officially in an AI bubble, at least in terms of valuations. We had big short memes all over Twitter, and more thoughtfully, Framework Ventures investor Van Spencer wrote, feel pretty comfortable calling the local top on AI as OpenAI goes out to sell shares at $90 billion. AI is likely frothier than crypto was at its peak. There simply isn't enough end user demand for AI-enabled products yet.

Rounds getting done at 100 to 1,000 X ARR. AI startups already starting to feel fatigued as incumbents simply adopt the tech and distribute it, themselves. Meanwhile, you can buy high growth at less than 20 PE in crypto. When the AI bubble pops, I'll be there to buy the capitulation. It's not personal. I believe AI will be hugely important over the long term. Now, there's an interesting question about whether Open AI being able to raise at $90 billion, and that being legitimate, and other AI startups being overvalued are actually mutually exclusive events. What I mean by that is that we've talked a lot, including up until yesterday, about how challenging the environment is for AI startups, given that the market is really prioritizing

either A, completely novel and differentiated foundation models, or B, existing trusted relationships around enterprise data, both of which put small startups at a pretty significant disadvantage. But even in the case that many of those smaller startups are currently overvalued, it doesn't necessarily mean that OpenAI at 90 billion is. Remember, OpenAI is playing at that big foundation model space and is at least for now still beating all the other giants, indeed forcing some like meta to adopt a totally different strategy in open source than they might otherwise have. A $90 billion valuation seems like a lot, unless you believe that OpenAI is destined to be a trillion-dollar company. Now, in terms of things that OpenAI might be doing that would get investors

excited, the information is also reporting that Johnny I, best known as the head of design at Apple during the golden age in which it introduced the iPod, the IMac, and so many other of the innovations that we use every day, has been talking with Sam Altman around an AI hardware project. Now, interestingly, apparently Masayoshi Sun, the soft bank CEO, has talked to both of these guys about the idea. As the information writes, it is not clear what the device would be or if they will decide to build it, but the two leaders who are friends have been discussing what new hardware for the age of AI could look like. It's also unclear whether any new device would be made by OpenAI or a new company. Now, of course, the question is whether a hardware device

would actually be value accretive for OpenAI or would just be a huge distraction. The DNA that it takes to build great hardware is different than it takes to build great software. Now, it's not to say that OpenAI doesn't have that DNA, but it certainly hasn't been tested. Then again, Johnny does. And so maybe the tie-up makes some amount of sense. Now, it's fair to say that probably the most hyped AI-related hardware device right now is called Humane. Humane is a wearable AI assistant that attempts to totally change the interface for how we interact with computers. It was demoed first at TED earlier this year and represents a full break with the phone as the device is completely standalone and doesn't need a phone or PC to interface with it. So some of the things in that initial

demo included summarizing emails, calendar invites, and messages from a busy workday, the device quote-unquote seeing a chocolate bar and giving the person doing the demo advice about whether or not he should eat it based on his dietary requirements. There's a sort of projection screen that puts information on nearby surfaces. And maybe the most discussed part of this was a translation demonstration in which the CEO who was doing the demo holds down a button on the device, says a sentence, and then hears it read back in his voice in a different language, in this case, French. Now, it's interesting because we're seeing that sort of functionality become quickly commodified, given that we've just had announcements from YouTube and Spotify about

that sort of AI translation, but put in the context of a wearable device, especially for people who are traveling, and that's obviously quite a game changer. Now, interesting to note, in 2020, it was Sam Altman himself who co-led a $30 million Series A in that company, Humane. He also participated in a $100 million series B in 2021 and a $100 million series C in 2023. Humane also appears to be building on top of OpenAIs technology. And this AI integrated wearable thing is definitely a trend I'm seeing more of. Just yesterday I saw a video from Woop, which is a wearable device where the CEO said, breaking, we've partnered with OpenAI to launch Woop Coach today, the most advanced generative AI feature ever to be released by a wearable.

Members can now ask Woop anything about their data and receive instant feedback. So for example, hey, Woop Coach, what's a great strength training workout for me today? I just sat in traffic, how do I chill out, etc., etc., etc. So chalk AI integrated wearables up to another trend that we will need to spend some time on. Now, lastly today when it comes to Open AI, we're going to look at a little joke that some people didn't think was very funny. Now for context on this, let's actually go back a couple days to rumors that we got over the weekend on Reddit around advanced open AI internal models that hadn't been released yet publicly.

One of the models the poster claimed was called Iraqis, and they said it, quote, far exceeds GPT4's capabilities and is really close to AGI. The poster goes on, honestly, if GPT4 was slightly conscious, then there are going to be a lot of ethical talks in that regard about this model, but Altman will want to view it as just a tool, even if there is a chance it is somehow actually conscious. Now, the mystery was increased by the fact that recently, a Twitter account known as Jimmy Apples had tweeted the quote, AGI has been achieved internally. In general, people probably wouldn't take this very seriously, except for the fact that all the way back

in April, the account was talking about a more advanced model inside OpenAI called Gobee, and now we've had a number of different independent verifiers say that such a model with that name exists. Heightening the mystery even more, Jimmy Apple's Twitter account no longer seems to exist. But just yesterday, Sam Altman responded to a thread on Reddit, a thread in fact noticing that Jimmy Apple's account had been deleted from Twitter slash X, and Altman said, AGI has been achieved internally. People raced to Twitter slash X to talk about it. Jan Pelig said, first comment in seven years, which eventually led to Sam editing

it and saying, edit, obviously this is just meming. Y'all have no chill. When AGI is achieved, it will not be announced with the Reddit comment. Now, some people saw this as an unbelievable troll. The AI safety memes account posted an image of it saying troll 100 and saying LMAO, Sam A commented from his real Reddit account AGI has been achieved internally and gave everyone a heart attack. Others took it much more seriously. Someone tweeted, I'm all for humor, but dude, this is seriously not funny. If the commander of US stratcom tweeted about incoming missiles as a joke, they'd be fired if not sent to jail. Eliezer Yudkowski said, my guess, open AI is trying to preemptively destroy the word AGI in order to destroy the possibility of coordinated political action about it.

And yet others once again thought that that concern was just silly. Renji at Brick Road 7 says Sam's joke has definitely revealed some ridiculously panicky and performative exaggerated paranoid character traits from the Dumer crowd. And just really, really rubbing it in, applied research resident at OpenAI Will Depew, posted an image created with the new Dolly 3, which hasn't yet been released to the public, of a set of paper clips flowing over the stairs at an office building with a caption, breaking news, Open AI offices seen overflowing with paper clips. The joke, of course, being that one of the famous scenarios that people use to describe how AI could end the human race is if a super-intelligent

AI is programmed with the goal of creating the most paper clips, it might break down all the other organisms on Earth in order to get more raw material for those paper clips. It's meant to be an example of how AI doesn't have to be malicious in order to still cause extremely problematic consequences for humanity. But now, at least for one person inside OpenAI, it is a meme to tease the Twitter crowd with. Now, moving away from the AGI internal joke, I do think that Sully Omar makes a great point when he tweets. Everyone seems to forget that OpenAI had GPT4 vision since March, likely earlier. That was seven months ago. Imagine what they're cooking up behind the scenes. The point being, while the capacities inside OpenAI might not yet be AGI, they are certainly

a lot more advanced than what we've seen so far, and as such probably operating an even more ethical gray space than what has been released currently. Anyways, guys, interesting times Open A.I continues to dominate mind share and product share in the space. I, of course, will do my best to keep you updated on all the important changes, but for now I appreciate you guys listening or watching as always. If you're enjoying this, I would so appreciate it if you would take the time to leave a rating or a review. But in either case, until next time, peace.