Aaron (@aaronbatilo) header
Aaron (@aaronbatilo) profile picture

Aaron

@aaronbatilo

Putting the ML in YAML. Eng @CoreWeave. Previously at @MicrosoftAI, @InflectionAI, @Cohere, @Color I like smash melee

 Colorado, USA
coreweave.engineer
Joined June 2014
  • Tweets4.1K
  • Following148
  • Followers833
354Media
Photo by Aaron (@aaronbatilo)Photo by Aaron (@aaronbatilo)Photo by Aaron (@aaronbatilo)Photo by Aaron (@aaronbatilo)Photo by Aaron (@aaronbatilo)Photo by Aaron (@aaronbatilo)Photo by Aaron (@aaronbatilo)Photo by Aaron (@aaronbatilo)Photo by Aaron (@aaronbatilo)Photo by Aaron (@aaronbatilo)
Beam’s model technical report is now available. Read the full report to learn about the methodology our team used to train Beam end-to-end in-house, here: reflection.ai/documents/beam…
20
68
32
473
69,164
Image from ☆ mitch (@mitchukage)
47
6,741
136
41,506
364,747
Aaron retweeted
mitra (@mitrajoy_) profile picture
mitra
@mitrajoy_Nov 3
instead of a sticker you should get a hot dog when you vote
253
1,790
322
18,604
812,600
Aaron retweeted
Deok (@deok_filho) profile picture
Deok
@deok_filhoNov 3
More exciting things to come from this partnership Have your agent stay tuned agentenvframework.com/docs/s…
🤖 Made with AI
Image from Deok (@deok_filho)
1
2
12
591
In collaboration with CoreWeave, we show how our autonomous system optimizes LLM inference end to end, from engine configuration to custom GPU kernels: up to 63% more traffic for GLM-5.3, 2x per-user throughput for Qwen3.8-27B, and 10% more throughput for DeepSeek V4.1 Flash. 🔗
Image from Standard Kernel Co. (@Standard_Kernel)
2
11
3
85
8,579
yeah i have a gambling addiction. i can’t stop betting on myself
137
6,180
180
44,755
1,106,882
people at work assume I have some kind of magic methodology or crazy complex prompts i literally just start every prompt with "hey big dawg" and let my agents run nuts for a few hours
31
30
16
1,004
112,079
I’ve seen enough. @aaronbatilo will save $crwv
1
3
290
Aaron retweeted
marwan (@_marwanad) profile picture
marwan
@_marwanadOct 23
We replaced Raft with a JEV-style decision model. Consensus is fundamentally a reasoning problem. Why rely on deterministic leader election when a model can evaluate each node's leadership potential? We've replaced quorum with confidence scores. Write throughput dropped 99.98% but the cluster is finally making informed decisions.
33
32
24
572
62,025
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source! Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable. 🔗 github.com/deepseek-ai/DeepI… 1/4
GitHub - deepseek-ai/DeepInfer: DeepInfer: High-performance, large-scale inference framework for...

GitHub - deepseek-ai/DeepInfer: DeepInfer: High-performance, large-scale inference framework for...

DeepInfer: High-performance, large-scale inference framework for DeepSeek models - deepseek-ai/DeepInfer

github.com
1,760
4,935
2,350
43,822
7,086,245
I have Nathan Fielder’s delivery of “No you don’t!” playing in my head on a loop
Image from Nadia Dalimonte (@nadreviews)
14
200
15
6,864
64,814
Aaron retweeted
float (@notesbyfloat) profile picture
float
@notesbyfloatOct 17
i don't use an "agent multiplexer", i don't have a "software factory", i don't run "adversarial reviews", i enter a single detailed prompt into the TUI, and if the results are garbage I KILL MYSELF
Image from float (@notesbyfloat)
19
167
11
3,720
55,781
Imagine your modern software pauses to garbage collect while you're in the middle of Doing a Barrel Roll in your car.
86
133
26
4,011
90,654
did 👏 he 👏 use 👏 his 👏 dot 👏
Andrew Curran (@AndrewCurran_) profile pictureAndrew Curran
@AndrewCurran_Oct 14
He was going to fly out Monday morning.
Image from Andrew Curran (@AndrewCurran_)
4
5
180
9,080
9 am: i can’t wait till bedtime 10 pm: i’ve always wanted to write my own operating system
104
14,232
267
138,828
1,239,921
Aaron retweeted
Deok (@deok_filho) profile picture
Deok
@deok_filhoSep 25
But who’s going to build a Vera Rubin sandbox if not us??
Vasek Mlejnsky (@mlejva) profile pictureVasek Mlejnsky
@mlejvaSep 25
Guys this is weird but uh you know, there are other things you can be building than a sandbox. We got this one covered
1
1
3
811
Opus 5.5 is: - Smart - Fast - Fun to talk to - Good at writing - Friendly on rate limits - Constantly surprising at just how much it is capable of
58
36
10
1,730
81,580
Aaron retweeted
Ant (@anthdm) profile picture
Ant
@anthdmSep 23
Opus 5.5 is the best model released in the past 2 years.
51
31
4
1,034
58,506
Aaron retweeted
Mads (@madsf88) profile picture
Mads
@madsf88Sep 21
going between any two points in london takes 54 mins
34
27
7
331
20,694
Aaron retweeted
Thariq (@trq212) profile picture
Thariq
@trq212Sep 18
We're adding support for AGENTS.md to Claude Code. Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md. You can toggle this behavior in /config.
2,049
2,628
2,522
31,278
5,640,655
Load more