Sabitlenmiş Tweet
Aaron
2.5K posts
Aaron
@aaronbatilo
Putting the ML in YAML. Eng @CoreWeave. Previously at @MicrosoftAI, @InflectionAI, @Cohere, @Color I like smash melee
Colorado, USA Katılım Haziran 2014
172 Takip Edilen319 Takipçiler
Because that shit is optimal, confirmed
son of nurhaci@aisingioron
Why does he have the Asian American accent
English
@aaronbatilo by listening to many things ive said or reading the book
English
I heard people needed clarification that my book was a post-training book posttrainingbook.com
English
Our inference stack, optimized for Blackwells, with a novel attention kernel and many new optimizations has started rolling out!
It's already charting on Artificial Analysis, eg: #1 speed and latency for @Kimi_Moonshot Kimi 2.6. #1 on latency on @MiniMax_AI, and miles ahead of other GPU endpoints.
artificialanalysis.ai/models/kimi-k2…
artificialanalysis.ai/models/minimax…
English
/goal in codex is like SOOOO much better. /goal in Claude Code ends up just giving up sometimes, like it literally just stops working while "Goal active" is just sitting there.
I think it has something to do with loop prompts that execute in the middle of goal completion
English
At what GPU cluster size do you stop being experiment limited and become gpu capacity limited
English
@charles_irl How do you conceptualize the difference with your idea versus dropout?
English
my gut says that to solve float numerics problems from nondeterminism x nonassociativity, we need to think bigger than determinism.
models could eg be trained with large amounts of "implementation noise" so that the learned network is more robust to implementation skew.
English
Let's start a cooking show where we try to get various ingredients to have the Maillard reaction. We'll call it:
Will It Brown?
kalomaze@kalomaze
bro
English
Aaron retweetledi
No matter the amount of full attention sparsity, decreased attention dim, MLA/GQA, infinite long context will always have to deal with quadratic prefill cost.
Kind of begs the question on if we can successfully train a model with local prefill and global decode?
English
This is like mang0 joining Team Liquid for people who know what GALINT is.
Andrej Karpathy@karpathy
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
English
Aaron retweetledi
