Mahesh Sathiamoorthy (@madiator) - Twitter Profili

Sabitlenmiş Tweet

Happy to finally make this announcement of our Seed and Series A raises. It's been a great journey since I left Google DeepMind 2.5 years ago with a goal to democratize post-training! Post-training gets easier if you have access to good data and now, RL environments. That's why we have poured our energy into doing data research. And we will continue to make Bespoke into one of the world's best data research labs. Thank you everyone for your support and thanks to the incredible Bespoke team that has done amazing work so far! PS1: Yes, we got busy with building after our Seed (MiniCheck, Curator, OpenThoughts, Terminal Bench..) so we didn't get a chance to announce the Seed raise! PS2: I will be at ICML starting Wed!

Bespoke Labs@bespokelabsai

We’re thrilled to announce a $40M investment that will fuel our mission to make AI agents reliable. For the past two years, we've been heads-down doing world-class data curation research and shipping best-in-class reinforcement learning environments for training and optimizing AI agents. This funding lets us go a lot deeper on both. Thank you to our investors @Wing_VC, @MayfieldFund, @8vc, @thehousefund and our angels such as Jeff Dean, Dheeraj Pandey, Tristan Handy, and several others from Anthropic, OpenAI, Meta. And thanks to the frontier labs and enterprises we work with every day, for sharing our vision for a future where agents can run autonomously for weeks and months at a time. (more below)

English

41

17

258

36.7K

Mahesh Sathiamoorthy@madiator·21h

This is cool. Curator allows you to call models on a lot of prompts to get the responses and also supports fine tuning with tinker etc. Should we combine those?

Sam Hogan 🇺🇸@samhogan

We're releasing Inference AutoTune Distill any frontier model into a 1-30B parameter task-specific SLM with only 25 lines of code automatically route requests to reduce cost and latency by >90% ~2 hours and <$250 to train. You own the weights Available in private beta today

English

3

0

13

3.5K

Mahesh Sathiamoorthy retweetledi

Sebastian Raschka@rasbt·2d

For agentic coding, one can say: - Unless you need Terra Ultra perf, it's always better to use a Luna model with higher effort setting (same or better performance but cheaper). - Forget everything below Sol High, use Luna with higher effort settings here - Forget Sol Extra High, use Terra Ultra here - The extra cost of Sol Ultra is probably not worth it over Max

English

323

444

4.5K

907.6K

Mahesh Sathiamoorthy@madiator·2d

In Gangnam, and i have seen more exotic cars in three days than my entire lifetime in Bay area (let's stop at "many years"). Several maybachs, rolls royce, lambo. There is this poor Bentley parked on the side of the street for many days..

English

3

0

21

2.6K

Mahesh Sathiamoorthy@madiator·2d

Congrats Alex and team!

Alex Dimakis@AlexGDimakis

Best Paper Award to our ICML Workshop paper 'Fantastic Adaptive Taxonomies and How to Use Them'. Let me summarize briefly what the paper is about: Failure taxonomies are becoming increasingly important. We show that failure taxonomies can be used in multiple ways: 1. As a test-time scaling tool for best-of-N judges, 2) as a mutation feedback mechanism in optimization loops and 3) as runtime feedback for coding agents. Previously people used MAST and other hand-made fixed failure taxonomies. In this paper we show how to create the failure taxonomy adaptively and dynamically: We observe agent rollouts and create an adaptive failure taxonomy, bespoke to the agent weaknesses and task challenges. These adaptive taxonomies give a massive boost: On Terminal Bench 2 we get 89.9% with Opus 4.6 / Forgecode harness and the adaptive taxonomy used with a Best-of-N Judge, outperforming fixed taxonomies by 15%. (1/n)

English

0

16

2.1K

Mahesh Sathiamoorthy retweetledi

will brown@willccbb·6d

@AlexGDimakis semi-related: i've found myself frequently referencing OpenThoughts lately as an explanation for why MOPD + same-family teacher OPD work much better than OPSD "QwQ is a better teacher than R1 for Qwen2.5-7B" is a very compelling result!

English

2

1

22

3.4K

Mahesh Sathiamoorthy

Keşfet