The AI Daily Brief: Artificial Intelligence News and Analysis - Are We In An AI Bubble?

Episode Date: June 8, 2023

Is AI in bubble territory yet? Ever since Nvidia's massive move up recently, people have been asking this question with renewed vigor. 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 exploring whether we're in an AI market bubble. Before that on the brief, OpenAI says they're not training GPT5. Google says Bard is 30% better. And yes, I'm losing my voice. The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Like, subscribe and share and go to Breakdown.network for more. Welcome back to the AI breakdown brief, all the AI headline news you need in five minutes or less. We kick off today with an update from Google's bar.

In the battle for supremacy when it comes to LLM-powered chatbots, Google has now announced that BARD is using a new technique that they're calling implicit code execution, which runs code in the background detecting computational prompts, and improves the accuracy of word and math problems by 30%. In their blog post, they explain, large language models are like prediction engines. When given a prompt, they generate a response by predicting what words are likely to come next. As a result, they've been extremely capable on language and creative tasks,

but weaker in areas like reasoning and math. In order to help solve more complex problems with advanced reasoning and logic capabilities, relying solely on LLM output isn't enough. Our new method allows BARD to generate and execute code to boost its reasoning and math abilities. This approach takes inspiration from a well-studied dichotomy in human intelligence,

notably covered in Daniel Kahneman's book Thinking Fast and Slow. Speaking of Google, their Deep Mind Division has also published new research in nature about something that they call AlphaDev, which is an AI system that uses reinforcement learning, and has discovered enhanced computer science algorithms. Now this is the latest in DeepMind's Alpha series of AIs,

which includes AlphaGo, which beat the world champion at Go. Originally, these AIs were trained around playing games, but they've now been repurposed for other tasks. DeepMind writes, Rather than refine existing algorithms, Alpha Dev started from scratch in a computer's assembly instructions. To train it, we built an assembly game where it's rewarded for sorting data efficiently and wins by finding a correct faster program.

This led, they say, to improved algorithms for sorting, sorting, which is obviously fundamental to things like social media algorithms, how information is processed on devices, and more. They also said that AlphaDev found ways to improve hashing, increasing the speed by 30%. Now, speaking of improved algorithms and better LLMs, OpenAI has claimed that it is still not training GPT5. This is something that Sam Altman said in his testimony before Congress a few weeks ago, and he reinforced it at a conference recently held by the Indian newspaper The Economic Times. Altman said, we have a lot of work to do before we start that model. We're working on the new ideas that we think we need for it, but we are certainly not close to it to start.

Altman also made news on this trip by saying that OpenAI has no plans to go public. The quote that everyone is running with is this one. When we develop superintelligence, we're likely to make some decisions that public market investors would view very strangely. The chance that we have to make a very strange decision someday is non-trivial. My read on this is that when it comes to really important questions, Sam doesn't want a fiduciary responsibility to increase profits at all costs. And while he says he doesn't want to be sued by the public market or Wall Street, that isn't stopping people from suing open AI in private right now. A radio host in Georgia is suing open AI after chat GPT told a

journalist that he was embezzling funds from a gun rights nonprofit. Gizmoto reports that it's maybe the first of its kind libel lawsuit, alleging that an AI damaged his reputation by making the claims. The DJ's attorney John Monroe said, quote, while research and development in AI is a worthwhile endeavor, it is irresponsible to unleash a system on the public knowing that it fabricates information that can cause harm. Speaking of causing harm, a number of senators aren't very happy that Meta's Lama AI model was leaked. Senator Richard Blumenthal, who hosted that hearing that Sam Altman appeared at a couple weeks ago, writes on Twitter, Meta released its advanced AI model Lama with seemingly little consideration and safeguards against misuse, a real risk of fraud,

privacy intrusions, and cybercrime. Senator Josh Holly and I are writing to Meta on the steps being taken to assess and prevent the abuse of Lama and other AI models. Now, one of the things that's interesting about this is that in that memo from a Google AI researcher that got so much attention a couple weeks ago, the researcher claimed that it was, in fact, the leak of Lama's full model that led to such an explosion of open source development. Given how angry these senators are, it shows the type of challenges the open source community might be in for when it comes to AI regulation. However, if senators Blumenthal and Holly want to use AI for themselves in the course of their

government work, they now might be able to. Microsoft has announced that they're bringing the GBT4 model to their Azure government cloud computing service, of which it counts as customers a number of U.S. government agencies. Federal customers of Azure include the Defense Department, the Energy Department, and NASA. Later today in Washington, British Prime Minister Rishi Sunuk will hold a meeting with President Joe Biden. Among other discussion points, Sunuk will invite Biden in the U.S. to participate in a UK-hosted Global Summit on AI Safety that is slated for this fall. A spokesperson said that the summit was, quote,

for like-minded countries who share the recognition that AI presents significant opportunities, but realize we need to make sure the right guardrails are in place. Now, when that same spokesperson was asked if it was aimed to counter China and Russia, they said, no, it's about looking at technology that is developing extremely quickly, perhaps faster than even those involved in its creation expected. And while they say that that meeting might not be about China, people are definitely taking note of advances in Chinese LLMs. Michael Frank shared, for example, a new Chinese LLM model that outperforms chat GPT 3.5

and Lama's 65 billion parameter model, but which isn't quite at GPT4's level yet. Finally, we close on a fun one. After weeks of waiting, Runway's Generation 2 text-to-video model is now available for everyone. They've been rolling out access to this in beta for a few weeks, but now they've opened it up to the wider world. If text-to-image allowed everyone to become a photographer or an artist, text-to-video might allow anyone to become a director. Already we're seeing artists experiment with it. Jared Lido published a video made using Runway with the latest. latest 30 seconds to Mars song in the background, and Weezer also created a tour promo with the

Generation 2 tool. It's highly likely that I do a full video about Gen 2 and text of video soon, so keep an eye out for that. But for now, that is it for the AI breakdown brief. If you're enjoying, please like, subscribe and share and click the notification button so you don't miss an episode, and I'll be back soon with the main AI breakdown. Are we in an AI bubble? And if we are, how bad is it?

Welcome back to the AI breakdown. When we talk about bubbles in technology, there are really two different possible dimensions of that. One is referring to the idea that the technology is all smoke and mirrors and not really advanced. The other is the idea that it's mispriced somehow by the markets. That market enthusiasm has outstripped the capacity of that technology to deliver value in the real world as it exists currently. Today we're talking about the market side of a potential AI bubble. For the last few weeks, smart analysts have argued both sides of this, that AI is overvalued, overhyped, overbought, but then at the same time, some have argued that we're still not understanding, just how

disruptive and transformative it is. This conversation heated up a couple of weeks ago when a really fascinating thing happened. For everything outside AI, the main discussion and what was shaping market action was the debt ceiling debate. House Speaker Kevin McCarthy and President Biden couldn't get it together, and we were careening into the U.S.'s first debt default in history. Still, as that happened, Wall Street wasn't in turmoil the same way that you might expect, and part of that was that NVIDIA was leading a rally the opposite direction. The CNN article starts, The debt-sealing debate and possibility that the United States could default on its financial obligations

have hung over the heads of investors for nearly six months now. But Wall Street appears to be largely shrugging off the ongoing negotiations as noise, instead opting to focus on a better than expected first quarter earning season. Now, of course, the Nvidia earnings weren't just good. They were spectacular, historic even by Wall Street standards. They revised up estimates of quarter one revenue and profit in a way that beat analysts' expectations, but even more, they blew current quarter estimations out of the water. Analysts had expected current quarter revenue of around $7 billion for Nvidia,

and they estimated it to be closer to $11 billion. The net impact of that was that Nvidia joined the rarefied error of the trillion-dollar valuation club. However, even before Nvidia, some were saying that we were in an AI bubble. Bank of America in a report called it a baby bubble that echoed the dot-com era. Michael Hartnett, the chief investment strategist at Bank of America Global Research, wrote in a note that bubbles, whether they're investing in the right things or the wrong things, are always started by easy money and are always ended by rate hikes. Hartnett pointed out that the speculative mania in the late 90s and early 2000s

was one of the driving reasons that the Fed restarted monetary tightening. Others were a little bit more measured in their bubble talk. Fundstrat, for example, said that they thought that mega-cap tech stock, were overbought and that their rally was close to stalling out, but that investors were right to be optimistic about AI. Others suggested that the rise of a stock like Invidia was creating a bubble by making people think that other tech stocks were going to follow along, even though Invidia was exceptional. Terese Pelletti wrote, Invidia increased the fervor for artificial intelligence on Wall Street, but plunging software stocks showed Thursday why betting that the chipmaker's AI revenue will

quickly ripple through the rest of the tech sector is a losing proposition. Terese pointed to C3.AI, which fell 14% after giving soft revenue guidance, crowd strike which fell 12%, sales force which lost 6%, and Okta which dropped 16%. Terese writes, those companies largely beat expectations both for results and guidance with their reports, but did not provide anywhere near the level of Nvidia's audacious forecast. Investors likely had outside expectations heading into the results after bidding up software stocks in the past month. Terese points to comments from an analyst regarding C3's results as emblematic of Wall Street's view more broadly. Jordan Berger and analyst at Third Bridge wrote, Despite AI optimism, we are hearing from our experts that generative AI is unlikely to represent

a significant revenue-generating opportunity for the company, and more likely represents a means for augmenting the user experience and searchability of the current platform. The company reported progress towards deploying generative AI technology within its platform, but how this will translate to revenue upside remains to be seen. As always in this type of situation, it pays to focus on the specifics, not just the generalities. The layer one basic analysis was invidia equals AI equals going up, so other companies that equal AI equal going up. But invidia is a unique company. They are the primary manufacturer of a key component of the entire AI movement, which is of course GPUs. Right now, the demand for GPUs radically outstrips the supply because of the rise of chat GPT, mid-journey,

stability AI, runway, and all of these other AI tools that are now entering the mainstream. Indeed, in a recent conversation with third-party developer partners, open AI's Sam Altman said that their roadmap was suffering because of this GPU shortage. Big fundamental features that they had wanted to bring to market this year, including a longer 32,000 token context window, as well as multidolity are in the waiting wings now because they simply can't get their hands enough GPUs. The point that this is about AI, but it isn't widely applicable to every company outside Nvidia. On top of simply learning the wrong examples from Nvidia, some say that people's focus on AI is making them completely ignore the larger macro environment. Contrarian Bear David Rosenberg

said that the price bubble in AI stocks could actually wreck the rally. No question, we have a price bubble, he said to CNBC's fast money last week. And in a recent note, he warned more broadly that the stock market rally probably doesn't have legs for long. Rosenberg wrote, There are breadth measures for the S&P 500 that are the worst since 1999. Just seven mega-caps have accounted for 90% of this year's price performance. You look at the tech waiting in the S&P 500, and it is up to 27%, where it was heading into 2000 as the dot-com bubble was peaking out, and soon to roll over

in spectacular fashion. Just today, Bloomberg published a widely shared piece called AI Stock Supercharge a Tech Bull Market, and maybe a bubble. The sudden frenzy over bots gives investors a powerful, if scary, new story to latch on to. Now, what this article points out is that there's more than just the traditional phomo dynamics going on right now. Bloomberg quotes Vincent DeLar, the director of global macro strategy at Stone X, who said, It's not just fomo. My impression is that it's F OBR, fear of being replaced.

Basically, we think, oh my God, AI is going to take over the world, my job. And the only way I can hedge it is by owning the damn robots. So you buy Nvidia. you buy Microsoft. Cam Harvey, a finance professor at Duke University, said, We've seen this before. Something gets hyped. Companies start to change their name so that you can inject AI directly into the name. AI is mentioned on conference calls, company websites, and public releases. It can go up very fast and come down very fast. This is not just talk. Consider the history of things that get hyped. That's exactly what happens. Mark Dow tweeted something similar,

saying all of a sudden every business has to be able to explain to their board what their AI strategy is. Asset managers are now going to have to do the same. If you don't have an AI strategy on your books, you're going to have to get one whatever the valuations. Still, not everyone is as convinced. Legendary investor Stanley Druckenmiller recently sat down with Bloomberg Invest and had this to say about AI and invidia. Do you think that all of AI makes it through this recession, or do you think that some areas of the market, particularly in AI, start to look like they're in bubble territory? Well, all of AI is not going to make it through whether we have a recession or not. because they haven't separated the wheat from the shaft yet.

But I do believe I think AI is real. It's probably, it could be as transformative as the Internet. It's a huge thing. And I think I've argued publicly that if staples can go up in price in a recession, why can't a company like NVIDIA, if they go up, if they go up, If their orders and earnings go up 70% in a hard landing, which is what I think would probably happening, it's not clear that it means that Embedia goes down despite the lofty valuation level.

If this is a secular move, if this thing is real, I mean, it's already making the top coders seven to eight times, seven to eight times more productive than they were five months ago. If it's as big as I think it is, Nvidia is something we're going to want to own for at least two or three years, not for 10 months. And maybe longer. Now, one thing that people seem to agree on is that AI from a bubble perspective is certainly less bubbly than crypto was just a couple years ago.

Nicholas Bonsack, the president of Stratiga's security, said, if I were to use past frenzies as a reference point, it's nothing like people saying to me two years ago, what coins should I buy? The crypto craze was on a whole other level. Wharton's Jeremy Siegel also said that it's not a bubble yet. He recognized that hype around AI plus Nvidia's earning statements maybe got the stock a little out of sync with reality,

but that it didn't constitute a bubble at present. He said, in the long term, I would say, Nvidia shares were probably slightly overvalued. But for the short term, we know momentum can carry stocks far higher than their fundamental value, and no one can predict how high they might go. Are you suggesting that these AI stocks are a bubble? No, they're not there yet.

And they got real worth. I mean, they got real earnings behind them. I'm not saying they're not eventually going to go too high. But I think, you know, I think we threw out a bubble way too easily. I mean, you know, to me a bubble is a stock that's four or five times above its fundamental values. And, you know, I don't think we're anywhere near that then. Others are putting some numbers behind this. One analyst Michael Goldstein has said that AI stock valuations just aren't near the level

yet that we've seen in past innovation bubbles. The managing partner of empirical research partners said, the relative forward PEs of today's AI leadership are still a far cry from what we've seen at the peaks of past innovation waves. In fact, they sit close to the level reached a year before the tops. Now, this might not be the case forever, Goldstein said. The combination of record-setting relative returns

and the stock's elevated arbitrage risk suggests it won't take too much to create a correction, but on the whole, the valuation of the stocks doesn't yet look excessive. Whether we're calling it a bubble or not, pretty clear that investors are piling into the space. Tech funds, for example, saw record inflows of 8.5 billion in the last week of May. But whenever we talk about market prices, it's important to differentiate short term from long term. Part of why AI is capturing so many people's imaginations

is that it seems like it could be the root of a productivity boom the likes we've not seen in a very long time. In May, Goldman Sachs senior strategist Ben Snyder told CNBC that over the next 10 years, AI could increase productivity by 1.5% per year. That Goldman believed could increase S&P 500 profits by 30% or more over the next decade. Just this week, Goldman wrote another note in which they said, We assume that widespread AI adoption occurs in 10 years and lifts trend real GDP growth by 1.1 percentage points for 10 years. In this scenario, earnings per share in 20 years would be 11% greater than our current assumption, and the S&P 500 fair value would be 9% higher than today, holding all else equal.

The analyst went on. However, a wide degree of uncertainty exists around the potential productivity boost and the ability of firms to translate AI into increased profits for 10 years. Based on a range of productivity scenarios, we estimate the benefit to S&P 500 fair value could be as small as 5% versus current levels and as large as 14%. Legendary hedge fund investor Paul Tudor Jones said something similar, saying that AI will spark the next productivity boom. He said, I think that the introduction of large language models, artificial intelligence, is going to create a productivity boost. that we've only seen a few times in the last 70 years. So how should you play it? Well, to be clear, this is not a financial advice podcast in any way. So I'll leave you with a couple of quotes from Twitter.

Tom Shaughnessy from Delphi Digital says, AI is definitely in a bubble. Same as every last tech breakthrough. The bubble attracts mass interest, but as humans, we always get ahead of ourselves. So should you stay on the sidelines? Absolutely not. Just be valuation sensitive and focus on long-term people.

Alex Good writes, If you treat AI as an investment bubble, you might be right. But when you approach something as a fad, you're keeping yourself from meaningfully engaging with it and using it as a tool. I agree with this. I think it's important to separate what you think about AI market prices and stock PE ratios from AI as a technology category. And of course, even within the technology category, there's no way that the vast majority of the venture capital pouring in right now is actually going to return. There will be big winners and far more losers. But the technology's entrance, into the mainstream is not just an accident, a fad, or hype cycle. People are using chat GPT because they recognize how much it can do for them. Anytime you have that sort of seismic shift, it's worth getting engaged. Anyways, guys, that is it for today's AI breakdown. I hope you found this useful. If you did, please like, subscribe and share. Click that notification button so you don't miss an episode. Go check out the podcast version. I would love for you to listen to this show or subscribe to the newsletter. All of that information is down below, and I appreciate wherever you consume this

content. Until next time, peace.