Charlie Snell
@sea_snell
Followers
8K
Following
74K
Media
529
Statuses
5K
PhD student @berkeley_ai; research @cursor_ai; prev @GoogleDeepMind. My friend told me to tweet more. I stare at my computer a lot and make things
Berkeley, CA
Joined April 2013
Can we predict emergent capabilities in GPT-N+1š using only GPT-N model checkpoints, which have random performance on the task? We propose a method for doing exactly this in our paper āPredicting Emergent Capabilities by Finetuningāš§µ
14
76
574
Weāre not playing that game. You shouldnāt have to register on someone elseās platform just to run your agentās payments. We shipped full x402 facilitator functions inside our SDK, so you can manage, execute, and verify gasless payments entirely on your own infra. No signups.
8
24
63
Tinker provides an abstraction layer that is the right one for post-training R&D -- it's the infrastructure I've always wanted. I'm excited to see what people build with it. "Civilization advances by extending the number of important operations which we can perform without
Introducing Tinker: a flexible API for fine-tuning language models. Write training loops in Python on your laptop; we'll run them on distributed GPUs. Private beta starts today. We can't wait to see what researchers and developers build with cutting-edge open models!
49
115
1K
Today, @ekindogus and I are excited to introduce @periodiclabs. Our goal is to create an AI scientist. Science works by conjecturing how the world might be, running experiments, and learning from the results. Intelligence is necessary, but not sufficient. New knowledge is
430
449
4K
Cursor can now control your browser. Agent can take screenshots, improve UI, and debug client issues. Try our early preview with Sonnet 4.5.
245
522
6K
this week we hosted the first build with cursor event where we gathered builders to build on top of background agents api and headless cli! here are all the demos from the night:
5
11
145
Capital-O Online RL
We've trained a new Tab model that is now the default in Cursor. This model makes 21% fewer suggestions than the previous model while having a 28% higher accept rate for the suggestions it makes. Learn more about how we improved Tab with online RL.
3
3
132
Holy shit theyāre doing on-policy RL by just deploying the model to prod lmao thatās so baller. also 2 hrs for a training step makes our 10 minute steps feel lightning fast @hamishivi ā¦ they probably have a bigger batch size though š
We've trained a new Tab model that is now the default in Cursor. This model makes 21% fewer suggestions than the previous model while having a 28% higher accept rate for the suggestions it makes. Learn more about how we improved Tab with online RL.
12
23
581
We've trained a new Tab model that is now the default in Cursor. This model makes 21% fewer suggestions than the previous model while having a 28% higher accept rate for the suggestions it makes. Learn more about how we improved Tab with online RL.
127
173
3K
Today Thinking Machines Lab is launching our research blog, Connectionism. Our first blog post is āDefeating Nondeterminism in LLM Inferenceā We believe that science is better when shared. Connectionism will cover topics as varied as our research is: from kernel numerics to
237
1K
8K
awesome to see another example of sleep-time compute deployed in production, especially on such a slick consumer app "setting up the memory can take up to 6 hours" congrats on the launch @nycintelligence š https://t.co/s8ZY8VVCc4
letta.com
Sleep-time compute is a new way to scale AI capabilities: letting models "think" during downtime. Instead of sitting idle between tasks, AI agents can now use their "sleep" time to process informat...
Cofounder is actually two agents - a memory agent which works in the background structuring a knowledge graph, and a real-time agent which accesses our memory. This allows us to take advantage of sleep-time-compute to better handle episodic memory and business ontology. In
1
7
57
yolo run summer is over scaling laws fall has arrived
1
1
63
Thereās something so whimsical about straight lines on log-log plots
1
0
20
first i thought scaling laws originated in OpenAI (2020) then i thought they came from Baidu (2017) now i am enlightened: Scaling Laws were first explored at Bell Labs (1993)
51
165
2K
