A look at the state of AI agents, the evolution of thinking models, the staggering need for inference compute in the coming years, automated research, and more
This was a great piece by Eric, and I'm in total agreement. The amount of intelligence consumption per capita is about to explode. We're going to be inference capacity limited for a long, long time. https://evjang.com/... [image]
@_sholtodouglas
Sholto Douglas on x
100% agree with his conclusions - Eric consistently predicts where the field is going.
@kevinroose
Kevin Roose on x
This is great, and the bits about inference demand ring true. It's trivially easy to generate 100ks of tokens in a single session with a coding agent (sometimes a single prompt!) and I'm not even using this stuff professionally. We're gonna need a lot more data centers.
@ericjang11
Eric Jang on x
While re-reading As We May Think I thought it would be fun to show some of the side-by-sides of inventions Vannevar Bush predicted in 1945 compared to their modern instantiation of information processing systems. 1/4 [image]
@aarthir
Aarthi Ramamurthy on x
“People who can direct teams of agents at goals and know how to judge what to focus on in a full-stack scope will experience an exhilarating level of productivity that makes making software a joy again. For roboticists: there is the age-old question of how much we should rely on
@deredleritt3r
Prinz on x
We are probably building too few data centers [image]
@nabeelqu
Nabeel S. Qureshi on x
Everyone underestimates just how much inference we will need. Excellent essay: [image]
@jenniferhli
Jennifer Li on x
Good read. And a powerful comparison. “Based on my own usage patterns, it's beginning to dawn on me how much inference compute we will need in the coming years. I don't think people have begun to fathom how much we will need. Even if you think you are AGI-pilled, I think you are
@altimor
Flo Crivello on x
Hard agree with this point. People underestimate how much inference we will need by many, many orders of magnitude. [image]
@gbrl_dick
Gabriel on x
incredible essay — the first half explains structural advances in LLMs very clearly, but i was struck by this towards the end. as models get better at longer horizon tasks, inference demand obviously goes up because longer reasoning uses more tokens but what i hadn't really [imag…
@ericjang11
Eric Jang on x
As Rocks May Think: an interactive essay on thinking models, automated research, and where I think they are headed. Enjoy! https://evjang.com/...
@deanwball
Dean W. Ball on x
Recently I did some back-of-the-envelope math and realized that my heightened use of coding agents in the last few months means that my daily token in/out consumption has probably increased by ~2 *orders of magnitude,* and I was a heavy user of reasoning chatbots before.