jay!

@jayz3nith

building self-improving agents before they build me

Los Angeles, CA

jayzenith.me

Joined March 2023

674 Following

85 Followers

jay!
@jayz3nith
32m
So I think the most interest aspect is if models will be able to obtain research taste through generalization via R&D envs. If so this will just be so insane. Like what a time to be alive.
Dwarkesh Patel
@dwarkesh_sp
Nov 13
The avalanche that hit math is coming for AI development. dwarkesh.com/p/rsi
jay!
@jayz3nith
Nov 10
i retract these first two sentences now. Just pondering on Fable 5.2 doing well in robotics, along with the bitter lesson, and thinking of continual learning algs that can update experts. I think we see whats going on. Last 2 sentences still legit though.
jay!
@jayz3nith
Nov 8
continual learning shouldn't be about remembering everything. Trying to keep a overly general model will be a constraint to continual learning. Harness memory is useful rn but it's really just better prompting. It's the gateway, the real thing has to happen in the NN
jay!
@jayz3nith
Nov 8
continual learning shouldn't be about remembering everything. Trying to keep a overly general model will be a constraint to continual learning. Harness memory is useful rn but it's really just better prompting. It's the gateway, the real thing has to happen in the NN
1
jay!
@jayz3nith
Nov 7
and so far seems like this is about continual learning through the harness which is just better prompting. But yeah will be an essential part of continual learning...
jay!
@jayz3nith
Nov 7
right to referencing Soren Kierkegaard, alright fully hooked in reading this now
jay!
@jayz3nith
Nov 7
right to referencing Soren Kierkegaard, alright fully hooked in reading this now
Anthropic
@AnthropicAI
Nov 6
Memory is what allows Claude to learn from experience. In a new Engineering Blog post, we share how “dreaming” helps Claude organize what it’s learned, correct its mistakes, and find what it needs more easily the next time. anthropic.com/research/compo…

jay!

jay!’s posts