AI: Kudos to 3 AI Arbiters on ‘Against the Grain’ Calls. ARD #91

Michael Parekh AI: Reset to Zero

4 hours ago

16 MIN READ

Today’s theme: kudos to three AI arbiters — the S&P 500, SemiAnalysis, and The Atlantic — for ‘against the grain’ calls this week. Two tie to next week’s SpaceX/xAI $1.75+ trillion mega-AI IPO, planning to raise north of $75 billion — the largest IPO ever. The third has to do with the tendency by AI companies and researchers to keep anthropomorphizing AI computing technologies. Three Takes today, each with my Take — and my Overall Take.

(1) S&P Holds the Line on SpaceX/xAI Early Index Inclusion

Bloomberg had it — “SpaceX, other mega IPOs denied fast index entry by S&P” — no waiver of the typical year of public-market seasoning, despite the pressure ahead of the IPO. Nasdaq went the other way, and will include SpaceX within 15 trading days. My “We are the Geese” frame on passive index force-feeding is on X, Elon’s boundless AI ambitions are in AI-RTZ #1066, and the mega-AI IPO filing wave is in ARD #81.

MP Take: S&P 500 is the largest of the index funds with over $10 trillion in passive funds and ETFs using it as their baseline. This means those funds won’t include SpaceX and other AI IPOs until a year of seasoning in the public markets — SpaceX gets the chance to prove or disprove the elements of its business model: xAI’s frontier-model business, Starlink’s core revenue engine, and the launch business in the middle. The move balances the interests of mainstream investors with the imperative of systematic and consistent processes for index inclusion. That needs to be applauded, even at a time of long-term AI enthusiasm. As a former head of Internet Research at Goldman in the nineties, these are calls I’m very familiar with — balancing the hopes and narratives of companies building the future against the reality and pace of what would actually be accomplished. Tough job. I’ve done it.

(2) SemiAnalysis Does a Balanced Assessment of ‘AI Data Centers in Space’

SemiAnalysis — Dylan Patel’s semiconductor research shop, whose work I’ve long followed — ran the deep read: “To Boldly Go: The Case for Space Datacenters.” Realistic parameters on how doable space datacenters are in the timeframes being talked about, and their pros and cons vs the terrestrial data centers where major tech companies are spending north of $2 trillion this year, next year, and the year after at least. The race for AI data centers in space is in AI-RTZ #936, SpaceX/xAI IPO prep in AI-RTZ #974, Tesla’s ‘Robotaxi & Robots’ ride in AI-RTZ #510, and “Bumpier Roads for Self-Driving Cars” from October 2023 here.

MP Take: Elon has consistently over-promised and under-delivered on major technology promises — starting with ‘fully self-driving cars,’ to ‘robotaxis,’ to ‘humanoid AI robotics’ like Optimus. AI Data Centers in Space is his latest promise with core unsolved technical challenges that will take years longer than promised — as judged by neutral, deeply technical analysis by scientists and researchers. The SemiAnalysis report takes a systematic approach to the technical and financial realities, and even though it may be optimistic in some places, the overall report lands more conservatively vs the promises laid out in the SpaceX/xAI mega-AI IPO filing. Much of the science doesn’t exist yet — it has to be invented, at far more money, time, and technical questions than is understood. A must-read for those curious about the pros and cons here.

(3) The Atlantic Calls Out AI Anthropomorphizing at Anthropic & Other Frontier LLM Labs

The Atlantic has the piece — “No, AI Is Not Conscious.” Alongside it — Anthropic’s own “When AI Builds Itself” and the WSJ on “Anthropic Urges Global Pause in AI Development, Flags ‘Self-Improvement’ Risk.” One of the clearest pieces on the illogic of treating AI as more than a computing technology — especially at a time when fear in the US is being propagated by some of the founders themselves. My “Don’t Anthropomorphize the AIs” from September 2023 is here, the accelerating costs of ‘humanizing’ AI in AI-RTZ #809, making AI look like it’s ‘thinking’ in AI-RTZ #630, and “On the Shoulders of Giants” (OTSOG) from December 2023 here.

MP Take: AI technologies, even with their exponential improvements ahead with further scaling, are still a probabilistic math-calculation-based computing technology. Yes, driven by extraordinary innovations in hardware and software to make these calculations occur at unprecedented levels. But they are mirrors to human life across millennia (”On the Shoulders of Giants”) — reflecting back probabilistic results from ever-expanding pools of data based on human thought and activity.

Humans have a natural proclivity to humanize everything — from pets to inanimate objects. AI technologies are uniquely susceptible to these emotional connections. A reality being leveraged by the AI companies — even as they’re building the core technologies. This Atlantic piece does a good job of laying it out.

The broader inclination of the AI industry to tout ‘self-improvement’ by AI is also an exaggerated claim. It points to the ongoing need of human processes to gate and govern AI code in business and consumer products and services.

MP OVERALL TAKE

Notice what all three arbiters have in common: they’re all speaking truth to power of different types — with balance.

MP Take: In a bull market this strong — with $200 billion+ to be raised by three mega-AI IPOs at multi-trillion-dollar valuations in the next few weeks and months, and trillions of passive money potentially being force-fed prematurely into these vehicles of growth before their maturity — these kinds of calls are important to call out as kudos. Markets and technologies compound best when analysts and organizations hold the line on process, on the physics, and the plain language. The AI Tech Wave I’ve been talking about for over 1,100 nonstop days is real and historic — everything I’ve done professionally in technology for three-plus decades is the dress rehearsal for what’s coming. But it needs to be done with the right balance of optimism on the possible and pragmatism around the probable. It’s precisely why the arbiters who check these narratives — their veracity and their probabilities — deserve our attention. Especially now, as these investments go from the private realm to the public realm.

Gadget AI — Meta Inches Forward to a ‘Name Tag’ App in AI Smart Glasses. Recording Lights Next?

Wired has it — “Meta silently adds face-recognition code for its Smart Glasses to millions of phones” — the ‘Name Tag’ capability inching toward Meta’s AI Smart Glasses, the category leader at 7 million+ units. Walk into a room and your glasses identify the people you’re seeing via their social IDs — “that was Jim, you met him at such and such.”

And The Verge has the recording-lights side — Joanna Stern’s piece on the cottage industry removing Meta AI Smart Glasses recording lights — dozens of vendors around the country who, for $100, will incapacitate the camera light that turns on white when recording. Influencers want it off for point-of-view videos — in a country where recording people without consent is illegal in half the states. Meta leaning in on AI Smart Glasses science projects is in AI-RTZ #849.

MP Take: This is an area that needs more consideration than current technology and business forces will allow. The issues here are ones the industry has struggled with since Google Glass and its ‘glasshole’ days over a decade ago. The answers on these face recognition and recording light questions will be important signals and guideposts to tougher issues to come with AI devices, gadgets and software ahead. And technology and business imperatives will always run ahead of the ‘right things to do’ — if prior tech waves are any guide.

Questions

Q1 — What is the one ‘creepy’ AI feature MP is looking forward to?

A Name Tag-like feature. I’m terrible at parties and business events — there are probably at least ten people in that group I should be connecting with, and I’m terrible with names. Something that reminds me to say hello to that person because you share an interest — that would be great. It’s the human-cognitive assist billionaires already pay assistants to whisper in their ears. But it needs industry and government unified regulatory parameters.

Q2 — What is the one ‘creepy’ AI feature MP is more wary about?

Always-on audio and video recording devices — not just phones, AirPods and glasses, but the name badges Microsoft is experimenting with via Qualcomm. I’m worried about all of this coming so fast that there’s a backlash — the Google Glass lesson. Over time people are adaptable and smart enough to figure out what’s good and what’s bad; if the good outweighs the bad, the thing passes. We just need to do it thoughtfully — with narratives not just driven by stock prices and the financial incentives of the companies. Again, needs unified regulatory parameters. But likely won’t get it.

Source Reading — For the Full Context

For the full context, see the canonical sources:

Shorts Clips from today

Clip 1 — AI Name Tags: Creepy or Cool?

Watch on YouTube Shorts

Meta is inching toward a ‘Name Tag’ feature in its AI Smart Glasses — glasses that identify the people in the room via their social IDs. I’m terrible at parties and business events: there are probably ten people in that group I should be connecting with, and I’m terrible with names. Billionaires pay assistants to whisper exactly this in their ears.

MP Take: If technology can provide that without intruding on the privacy of people around you, that’s a good thing — the kind of human cognitive assist computers should give us. But it needs industry and government unified regulatory parameters.

Clip 2 — AI: Still a Computer

Watch on YouTube Shorts

The Atlantic’s “No, AI Is Not Conscious” is one of the clearest pieces on the illogic of treating AI as more than a computing technology. Even at the levels AI will scale exponentially in the next two or three years, it’s still a computer technology — probabilistic instead of the deterministic computing we’ve used for 70+ years, so it feels almost human. But it’s still a computer.

MP Take: That distinction matters as we assess the investability of these things, their applicability, how we govern them, and how we should fear them.

Clip 3 — No, AI Is Not Conscious

Watch on YouTube Shorts

The industry — Anthropic and OpenAI in particular — has tended to internalize the idea that AI is greater than a computer technology: that it’s on a human-equal pathway, that it will essentially build itself. The Atlantic lays out the illogic of that proposition better than anything I’ve seen.

MP Take: Don’t humanize AI technologies — these are just technologies. Humans have a natural proclivity to humanize everything, and AI is uniquely susceptible to those emotional connections. A reality being leveraged by the AI companies, even as they build the core technologies.

Clip 4 — Balancing AI Tech & Society

Watch on YouTube Shorts

Society is getting trained on what’s possible with AI — from search bots to agents to a lot more. How should we trust these devices? In half the US states it’s illegal to record people without their consent — and there’s now a cottage industry charging $100 to disable the recording light on Meta’s smart glasses so influencers can record point-of-view videos without anyone knowing.

MP Take: Technology and business forces will push these things as fast as they can — that has to be balanced. I’m not a regulatory guy, but there’s a way to consensually figure out that balance as an industry. These questions go far beyond the technology making them possible.

AI Ramblings Daily on AI-RTZ is here to think through AI and reset. Together.

Today’s AI-RTZ #1108 — Taiwan Semi, ‘Fed of the Global Chip Economy’, Speaks — TSMC sets a 30% growth band on chip production, the number behind everything from Nvidia to Apple and the trillions in AI investments we discussed today.

Tomorrow — the Saturday weekend Roundup, AI-RTZ #1109. Have a great weekend.

Thanks for joining us, AI Curious Folk. Stay tuned.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here.)

Links

Theme — 3 AI Arbiters

Bloomberg — SpaceX, other mega IPOs denied fast index entry by S&P: https://bit.ly/4e8Uruk
SemiAnalysis — To Boldly Go: The Case for Space Datacenters: