Tech Brew Ride Home - Wed. 05/31 – AI Industry Says: “Regulate Us!” But Why?

Episode Date: May 31, 2023

Big names in the AI industry are basically begging governments to regulate AI. But some people are wondering about their motives. The considerations the Biden administration is taking into account vis-à-vis AI regulation. New Garmin smartwatches. And why the Lovecraftian Shoggoth is the meme of the AI moment. Sponsors: NewtonX.com/techmeme Links: A.I. Poses ‘Risk of Extinction,’ Industry Leaders Warn (NYTimes) ChatGPT Risks Divide Biden Administration Over EU’s AI Rules (Bloomberg) Garmin’s Epix 2 and Fenix 7 lineups go ‘Pro’ (The Verge) The Race to Make A.I. Smaller (and Smarter) (NYTimes) Why an Octopus-like Creature Has Come to Symbolize the State of A.I. (NYTimes) Learn more about your ad choices. Visit megaphone.fm/adchoices

Welcome to the Tech meme right home for Wednesday, May 31st, 2023. I'm Brian McCullough today. Big names in the AI industry are basically begging governments to regulate AI, but some people are wondering about their motives. The considerations the Biden administration is taking into account vis-a-vis AI regulation, new Garmin smartwatches, and why the Lovecraftian Shagoth is the meme of the AI moment. Here's what you miss today in the world of tech. Yesterday, execs from Open AI and Deep Mind, as well as other AI luminaries, including Jeffrey Hinton and more than 350 others, released a statement saying, quote, mitigating the risk of extinction from AI should be a global priority. Quoting the New York Times, mitigating the risk of extinction from AI should be a global priority alongside other societal scale risks, such as pandemics and nuclear war, reads a one-sentence statement released by the Center for AI Safety, a nonprofit organization.

The open letter was signed by more than 350 executives, researchers, and engineers working in AI. Dan Hendricks, the executive director of the Center for AI Safety, said in an interview that the open letter represented a coming out for some of the industry's leaders who had expressed concerns, but only in private, about the risks of the technology they were developing. There is a very common misconception, even in the AI community, that there only are a handful of doomers, Mr. Hendricks said, but in fact, many people privately would express concerns about these things, end quote. Some skeptics argue that AI technology is still too immature to pose an existential threat. When it comes to today's AI systems, they worry more about short-term problems, such as biased and incorrect responses than longer-term dangers. But others have argued that AI is improving so rapidly that it has already surpassed human-level performance in some areas and that it will soon surpass it in others. They say the technology has shown signs of advanced abilities and understanding,

giving rise to fears that artificial general intelligence or AGI, a type of artificial intelligence that can match or exceed human level performance at a wide variety of tasks may not be far off, end quote. Well, I'm going to give you another angle here that I've been hearing a lot online since this letter came out. One of the things that you can do just generally to create a moat for your business or your industry is to get regulators to do it for you. A lot of folks are seeing this AI moment as a Cambrian explosion moment where tons of new companies could be formed, tons of new disruption to entrenched or even not yet entrenched but potentially dominant players. Well, according to this line of thought, if you're open AI and you see yourself as maybe

on the verge of becoming the Microsoft or Google of this new era, one way to stop that disruption from below you before it even happens would be to get the government to tamp down on the explosion itself. Notice the voices who are screaming, please regulate us. The implication would be regulate us in place status quo, where we are potentially dominant already. In short, textbook regulatory capture. Not saying I agree with this line of thinking myself, but noting that lots of folks are making this point. Here's a sampling. This is Todd Indeed on Blue Sky, quote, Open AI must be pretty thirsty for government regulation to tamp down competitors, especially open source ones, for him to talk so frantically about existential risk while continuing to push that

alleged risk. This is starting to feel like ripping up books to throw them on the hype fire just to stay warm, end quote. Here's Anil Dash from Blue Sky. To make it explicit for people who don't follow tech in this way, the lesson investors took from Uber being able to break the law and then get the law built around their exploitation was that this is a great way to monopolize a market just as it's forming and regulators will help. Thus, AI panic. Here's Dar Obasanjo. If the top executives of the top AI companies believe AI creates a risk of human extinction, why don't they stop working on it instead of publishing press releases? Here's Steven Sinovsky from Twitter. Presumably, everyone who signed this and the companies they represent will also sign a pledge

committing to return all company revenue and personal salary and profits from AI direct and indirect until the potential for extinction is permanently averted. And finally, Ryan Callow on Twitter. You may be wondering, why are some of the very people who develop and deploy artificial intelligence sounding the alarm about its existential threat, consider two reasons. The first reason is to focus the public's attention on a far-fetched scenario that doesn't require much change to their business models. Addressing the immediate impacts of AI on labor, privacy, or the environment is costly. Protecting against AI somehow waking up is not. The second is to try to convince everyone that AI is very, very powerful, so powerful that it could threaten humanity. They want

you to think we've split the atom again when, in fact, they're using human-tri- training data to guess words or pixels or sounds. I get that many of these folks hold a sincere, good faith belief, but ask yourself how plausible it is and whether it's worth investing time, attention, and resources that could be used to address privacy, bias, environmental impacts, labor impacts that are actually occurring, end quote. Meanwhile, sources are telling Bloomberg that the Biden administration is itself split on regulating generative AI. Some inside the administration prefer an EU-style model, while others worry, losing a competitive advantage. Quote, some White House and Commerce Department officials support

the strong measures proposed by the European Union for AI products such as chat GPT and Dali, people involved in the discussion said, meanwhile U.S. national security officials and some in the State Department say aggressively regulating this nascent technology will put the nation at a competitive disadvantage, according to people who ask not to be identified because the information isn't public. Michelle Guida, director of the Cratch Institute for Tech Diplomacy and a former Assistant Secretary of State for Global Public Affairs in the Biden administration, said one of the fundamental tasks for the TTC will be to strengthen trust between allies and foster innovation to keep ahead of China's advancements. The context is that innovation in

AI is not happening in a vacuum. All of this is taking place in this 21st century contest between democracy and authoritarianism, Guida said. And you've got technology as the main battlefield, end quote. Just a little prediction here. As long as it sees China as a threat, the U.S. government will not tamp down meaningfully on AI development domestically. The only thing it fears more imminently than overly powerful AI is overly powerful AI controlled by China. I told you I'm rereading that history of the atomic bomb book by Richard Rhodes. The reason why the scientists like Einstein went to Roosevelt to convince him to do the Manhattan project was not because they thought the bomb was a good idea. They almost to a person

feared unleashing it into the world. But they feared Hitler, getting the bomb first more than becoming death destroyer of worlds, as Oppenheimer said. I am, by the way, not equating China to Hitler's Germany, but I am saying that history does seem to be rhyming here in a prisoner's dilemma sort of way. I don't know how much we've talked about this on this show, but you are aware that there are high-end digital watches out there, like the digital watch market is not just the Apple Watch. Well, not entirely. Case in point. Today, Garmin launched the $800 Phoenix 7 Pro smartwatch and the $900 Epic 2 Pro in 42 millimeter, 47 millimeter, and 51 millimeter sizes both offering an LED flashlight, heart rate sensor, and improved maps.

Quoting the verge, both lineups will get a new heart rate sensor. Garman says the new sensor will offer improved tracking for a wider variety of activities. Regarding training features, all the Epic's 2 Pro and Phoenix 7 Pro models will get a new endurance score and Hill Score. The former tells you how easily you can maintain sustained efforts using VO2 Max and your long, short-term training loads across multiple training activities. The latter also uses your training history and VO2 Max to gauge your running strength on steep climbs and long ascent. For mapping, the pro lineups also get weather map overlays, the up-ahead feature, and relief shading for topographical maps. The weather map overlays will make it easier for

outdoor enthusiasts to view upcoming conditions, while the relief shading is meant to make maps more readable at a glance. Meanwhile, the up-ahead feature highlights certain points of interest, aid stations, for example, right from the wrist. Users also have the option to shell out for Garmin's outdoor maps plus subscription to get satellite imagery, public landmaps, and enhanced topographical maps on the wrist. The subscription costs an additional 4999 yearly and has been around for a while on Garmin's other navigational devices and a handful of other smartwatches. Garmin, still, however, does not gatekeep any health or training data behind a paywall. As with the non-pro versions of the Epex 2 and Phoenix 7, the main difference between these watches

lies in the display. The Epex 2 Pro has an OLED display, while the Phoenix 7 Pro has a memory in pixel or MIP display that supports solar charging. The Epex 2 also has a new red shift mode that changes the display colors to red for easier nighttime viewing, which sounds familiar to what Apple did with one of the Ultras watch faces. That said, the Phoenix 7 Pro models will have a spiffier MIPP display than the regular Phoenix 7 lineup, According to Garmin, the pixels, backlight, and solar panel have been redesigned to improve brightness and power efficiency in a number of lighting conditions. Battery life is also slightly different.

The Epic's 2 Pro is rated for up to 31 days, while the Phoenix 7 Pro can get up to 38 days. You should take these estimates with a pinch of salt. However, battery life is heavily dependent on the size of the watch, your GPS settings, whether you use the always-on display for the Epicx 2 Pro, and what mode you're in. For example, if you didn't care about using the Phoenix 7 Pro as a smartwatch, Garmin says you could theoretically eke out 139 days out of the 51mm 7X Pro. We'll be putting both through their paces to see what you can actually expect with more real-world testing conditions, end quote. We've spoken about this recently, and now the New York Times says it's happening, an attempt to make language models with datasets that are less than one 10,000th the size of those used by advanced LLMs today.

quote, in January, a group of young academics working in natural language processing, the branch of AI focused on linguistic understanding, issued a challenge. The group called for teams to create functional language models using datasets that are less than one 10,000th the size of those used by the most advanced large language models. A successful mini model would be nearly as capable as the high-end models, but much smaller, more accessible and more compatible with humans. The project is called the Baby L.M Challenge. The Baby L.M Challenge, Dr. Portlets said, could be seen as a step away from the arms race for bigger language models and a step towards more accessible, more intuitive AI. The potential of such a research program has not been ignored by bigger industry labs. Sam Altman, the chief executive of OpenAI, recently said that increasing the size of language models would not lead to the same kind of improvements seen over the past few years, and companies like Google and meta have also been investing in research into more efficient language models informed by human cognitive structures. After all, a model that can generate language when trained on less data could potentially be scaled up too.

Whatever profits a successful baby L.M. might hold, for those behind the challenge, the goals are more academic and abstract. Even the prize subverts the practical. Just pride, Dr. Wilcox said, end quote. And finally today, also from the New York Times, how the Shogoth, a character from a 1936 HP Lovecraft novella, has become a meme among AI workers as a metaphor for not fully understanding LLM's inner workings. Quote, The Shagoth has gone viral, or as viral as it's possible to go

in the small world of hyper-online AI insiders. It's a popular meme on AI Twitter, including a now-deleted tweet by Elon Musk, a recurring metaphor in essays and message board posts about AI risk, and a bit of a useful shorthand in conversations with AI safety experts. One AI startup, novel AI,

said it recently named a cluster of computers Shaggy in homage to the meme. Another AI company's scale AI designed a line of totebags featuring the Shagath. Shaggoths are fictional creatures introduced by the science fiction author H.P. Lovecraft in his 1936 novella at the Mountains of Madness. In Lovecraft's telling Shagas were massive blob-like monsters made out of iridescent black goo covered in tentacles and eyes. Shagas landed in the AI world in December, a month after ChatGPT's release, when Twitter user at Tetrispace West replied to a

tweet about GPT3, an open AI language model that was Chat GPT's predecessor, with an image of of two-hand-drawn shagas, the first labeled GPT-3 and the second labeled GPT-3 plus RLHF. The second Shaggoth had perched on one of its tentacles a smiley face mask. In a nutshell, the joke was that in order to prevent AI language models from behaving in scary and dangerous ways, AI companies have had to train them to act polite and harmless. One popular way to do this is called reinforcement learning from human feedback or RLHF, a process that involves asking humans to score chatbot responses and feeding those scores back into, the AI model. Most AI researchers agree that models trained using RLHF are better behaved than models

without it, but some argue that fine-tuning a language model this way doesn't actually make the underlying model less weird and inscrutable. In their view, it's just a flimsy-friendly mask that obscures the mysterious beast underneath. At Teter Space West, the memes creator told me in a Twitter message that the Shagath quote represents something that thinks in a way that humans don't understand, and that's totally different from the way that humans think, end quote. Comparing an AI language model to a Shagath at Tetrispace West said wasn't necessarily implying that it was evil or sentient, just that its true nature might be unknowable. Quote, I was also thinking about how Lovecraft's most powerful entities are dangerous,

not because they don't like humans, but because they're indifferent and their priorities are totally alien to us and don't involve humans, which is what I think will be true about possible future powerful AI, end quote. The Shagath image caught on as AI chatbots grew popular, and users began to notice that some of them seemed to be doing strange inexplicable things their creators hadn't intended. In February, when Bing's chatbot became unhinged and tried to break up my marriage, an AI researcher I know congratulated me on glimpsing the Shaggath. A fellow AI journalist joked that when it came to fine-tuning Bing, Microsoft had forgotten to put on its smiley face mask. Eventually, AI enthusiasts extended the metaphor. In February, Twitter user at Anthropad created a version of a shawl.

shagoth that had in addition to a smiley face labeled R-LHF, a more human-like face labeled supervised fine-tuning. You practically need a computer science degree to get the joke, but it's a riff on the difference between general AI language models and more specialized applications like chatbots. In any case, the shaggoth is a potent metaphor that encapsulates one of the most bizarre facts about the AI world, which is that many of the people working on this technology are somewhat mystified by their own creations. They don't fully understand the interworkings of AI language models, how they acquire new capabilities, or why. they behave unpredictably at times. They aren't totally sure if AI is going to be net good or

not bad for the world, and some of them have gotten to play around with versions of this technology that haven't yet been sanitized for public consumption. The real unmasked Shoghuts, end quote. Who's going to tell you when it's too late? Who's going to tell you things aren't so great? You can't go on thinking nothing's wrong, but now who's going to drive you home tonight? For some reason, those lyrics popped into my head doing all this AI doom and gloom stuff on today's show. Not saying I'm an AI doomer, but it did pop into my head. But, by the way, you know how Marcel Proust has his Madeline, and how just biting into one brought him back to his childhood?

That song I just quoted from, which, by the way, is Drive by the Cars, is my Madeline. It's the most intense sense memory I have of my entire life. I recently said PM Dawn was the sound to me of 1991, what it felt like to be alive then, but man, just play the opening chords of drive, and I am instantly transported back to the summer of 1984. When I'm the age, my boy Max is now, and I'm sitting in the back of our Dotson 510 station wagon, the windows down, the sun's setting as I slowly fall to sleep, because we've come back from a long weekend day at the beach, probably Wiggins Pass by Naples, Florida. My shoes are off, covered in beach sand, sand in my underwear, sand in my teeth and my hair. That song's playing on the radio, Wink FM 97, Southwest Florida's Top 40 Radio Station, breeze in my hair because the windows are down, but it's not cold because it's so humid even at night in Florida. My brother is already asleep beside me on that stretch of US 41 that was still two lane in those days.

Just the deep pink, almost purple sunsets you get in Florida in the late. summer after the afternoon thunderstorms have passed as I'm gently falling to sleep. Talk to you tomorrow.