Gradient Ascendant is a newsletter about AI/ML research, applications, prospects, weirdness, cultural impact, and history, highly opinionated yet reasonably rigorous, because AI is — by far — today’s most interesting technology wavefront. New posts/series are planned for every two weeks; subscribe away.

My background is here; briefly, I'm an engineer, journalist, CTO, archivist, and author. (My science fiction novel Exadelic will be published in September by Tor Books. It too is about AI.) Only a few years ago I would have resisted using the term artificial intelligence at all; machine learning felt far more accurate. I'm launching Gradient Ascendant because, well, I no longer think that.

Whatever you call it, the field is full of amazing breakthroughs. OpenAI's GPT-4 and DALL-E, with competitors such as Google's Imagen and MusicLM, have gone beyond recognizing patterns and transferring styles — the previous state of the art — to interpreting language well enough to create legitimately new and extraordinary works. Even this pales next to the accomplishments of DeepMind, whose AlphaFold and AlphaTensor have broken new ground in science and math. This wave of innovation has finally legitimized the term AI.

Of course we’re still not even remotely close to artificial general intelligence, or AGI, loosely defined as AI at least as flexibly smart as us. But a remarkable recent survey of AI practitioners revealed that a majority believe "AGI is a real thing that might be important: 58% agreed AGI should be an area of concern ... 57% said recent research has advanced us toward AGI."

I plan to write about large language models, transformers, and how diffusion models work; how people strive to cleanse AI model outputs of the bigotry too often inherent in their data; the secret languages of AI image generators, and the terrifying "Loab" and "Chungus" figures who seem to haunt them; why DALL-E can't spell, but Imagen can; the care and training of humanoid robots; why the future is on the side with the big compute; how data is like fast food; and many/any other AI-related subjects. As always, I'll be writing in large part to figure things out myself, and I’ll doubtless make mistakes along the way; sorry in advance. But sign up and I’ll do my best to help us figure out this strange and brave new world together.

Subscribe to Gradient Ascendant

AI/ML research, applications, prospects, weirdness, cultural impact, history, etc.


Jon Evans

Novelist (EXADELIC from Tor Books), engineer (Metaculus), journalist (Wired, The Guardian, The Walrus.) Previously: director GitHub Archive Program, CTO HappyFunCorp, columnist TechCrunch, tech architect See .