Discussion about this post

User's avatar
Porlock's avatar

Brilliant, but that hardly needs saying. But the current reaction to LLCMs reminds me of the one course in AI that I took at Berkeley, in the new Computer Science department (Dept of Electrical Engineering, their first affirmation that Computer Science was a thing). One of the subjects covered was a new idea which was then being abandoned: Artificial Neural Networks (a term I had to look up, praise be to the Internet, having forgotten it after all these years) which had briefly looked like huge advance when implemented in software. Hadn't worked, of course. Seeing this, I thought of Moore's Law, an unscientific thing that no one believes any more, which why it is being cited all the time with due apologies. So I applied it to the time from 1966 to the present.

Most interesting. Makes a software guy really appreciate hardware.

Expand full comment
Alex Tolley's avatar

I have a few comments.

1. "AI" as a term has been in common parlance since the 1950s, at least. It is an overarching term for the many techniques that have been used, including ML.

2. Intelligence is not well defined. Usually more operational and descriptive, but not carefully defined. Therefore, it is bandied about from academics to the "Man on the Clapham Omnibus."

3. LLMs have reached the stage where modern humans can anthropomorphise the LLMs much as the ancients did about natural phenomena. We are wired that way. In a fairy tale, we have substituted talking animals for a talking box.

4. The pouring of capital into infrastructure is the same as any other "mania", tulips, South Sea, etc. Even "smart people" convince themselves that there will be a return. We have had the phenomenon in railways, the internet, and fiber buildout. If anything, I would encourage this, as long as we can ring-fence the economic fallout when the bubble bursts. As in the Gilded Age, super-wealthy people are running the show. Maybe the fallout will persuade us to limit the wealth disparity that can lead to such hubris.

5. NVIDIA is selling shovels to gold mining companies. Cynically, the company may be selling the dream of "Gold in them thar hills" by salting the rivers with a few nuggets. More likely NVIDIA believes the hype and will support it as long as they can keep selling high-end GPUs. After all, it just did a revenue-sharing deal with POTUS in exchange to selling to China. Maybe that sucked in POTUS too?

6. Egos. At this point, it must be clear that this is a bubble that will burst. Huge egos cannot allow themselves to exit and admit they were mistaken. So the merry-go-round continues, until there is a "winner," perhaps selling expensive services to governments to generate some sort of hoped-for ROI.

My hope is that there will be enough value in the work to democratize MAMLM use in edge devices. [See Pete Warden's blog https://petewarden.com/2025/10/02/how-to-try-chromes-hidden-ai-model/ on accessing Google's NanoLLM in the Chrome browser. At a minimum, there will be talking stuffed animals, like a super Teddy Ruxpin, as in Aldiss's "Super Toys last All Summer Long" (filmed as A.I.)]

If A.I. is a con, it is some people conning themselves, in an infomedia system that is largely just cheerleading. [IIRC, Hitler was duped by Von Braun into diverting huge resources into building vengeance rockets that did nothing to win the war. In some respects, he did it again with the "Space race" that briefly put flags and footprints on the Moon, but did increase US global standing. Maybe a sort of AGI will do the same?]

Expand full comment
19 more comments...

No posts