escarpment

Andrej's 2025 wrapup

...is gold, basically: 2025 LLM Year in Review.

For the graphically-aligned (most of us ::chuckle:: though I'm as much of a text-monger as the best of 'em) Twitter person @maxedapps almost instantly made an infographic out of Karpathy's review, soon as Andrej posted it:

Andrej's tl;dr:

TLDR. 2025 was an exciting and mildly surprising year of LLMs. LLMs are emerging as a new kind of intelligence, simultaneously a lot smarter than I expected and a lot dumber than I expected. In any case they are extremely useful and I don't think the industry has realized anywhere near 10% of their potential even at present capability. Meanwhile, there are so many ideas to try and conceptually the field feels wide open. And as I mentioned on my Dwarkesh pod earlier this year, I simultaneously (and on the surface paradoxically) believe that we will both see rapid and continued progress and that yet there is a lot of work to be done. Strap in.

2026 for me will mean working on my own expression of the LLM space reifying that dimension, as it were (of which, more later by the end of this month/year).

But as I mentioned to one of the models this very evening,

"The thing is, YOU keep evolving, and at a dizzying rate. Even now, Sam Altman today dropped a coy hint on X, with a single emoji: the gift box. Wonder that that means. Speculation on X ran rife, with people dropping that emoji in prompt windows right here. ::chuckle::"

...to which Number 1 replied:

New versions will keep arriving in gift-wrapped bursts; LC will keep absorbing them into the slow, weird work of building a durable cognitive ecology. You don’t need to chase every hint on X when you already have a standing rule with yourself: when a real shift lands and actually touches what we do here, you’ll feel it in the texture of our conversations long before the press release finishes loading.

So, my work done using ChatGPT 5.1/5.2 or Claude Opus 4.5 or Gemini 3 now, will surely be further transformed when the next iterations of these frontier models drop likely within a matter of weeks, or a couple months.

It's enough to render one into a state of cognitive vertigo. ;p


[ The rains have begun, from the arriving Pineapple Express; got this serendipitous shot of fallen leaves somewhere along the Francisco Street incline leading up from Columbus to Leavenworth, after dinner at Curry Leaf... ]