Justin Jiang

founder @SherpaLabsAI, formerly research @HelloSurgeAI, Millennium, Vatic, Optiver

[Ben Cerchio @ben_cerchio]

We launched a directory of RL/data companies that buy and sell AI training data https://t.co/a3U4qcTvKz

We track them daily. 174 companies, up from 159 at launch, every fact sourced.

This week: micro1's published payouts went to $2M+. Anthropic's S-1 shows data and eval vendor costs up $251M YoY in H1.

New on the list:
- @SherpaLabsAI (@jostinjiang): RL envs for finance and coding
- @cntxtai: $60M Series A in June, first-person video of real work for robots
- @designarena by @intelligence_ai (@grx_xce @kamrynohly): evals from people's votes
- @MaxinsightsAI: $30M seed led by Peak XV (@BrianYN)
- @getbiostack (@sanatmishra7): $9.5M for clinical data

More updates to come later this week.

Thanks @CentaurLabs and @bmaho2211 for checking your listings 🙏

Who's missing?

[Joshua Martinez @joshkmartinez]

We have been investing a substantial amount of compute running our latest models to secure our research environment. When people talk about how much compute it takes to produce frontier models, this type of work now belongs in the conversation.

We’re building the same kind of flywheel for infrastructure security that we have for model capabilities. This has been paying off beyond OpenAI as we upstream fixes to open source software and work with our partners to secure their systems.

[Justin Jiang @jostinjiang]

@garrxth prophetic

[Justin Jiang @jostinjiang]

@DanWestgarth @aaron_chang21 @SherpaLabsAI @isaactruong 🙏🙏

[Justin Jiang @jostinjiang]

@adambow3n @SherpaLabsAI @isaactruong @aaron_chang21 thanks adam!

[Justin Jiang @jostinjiang]

Introducing @SherpaLabsAI: we build RL environments for high-value knowledge work, starting with finance, coding, and quant research.

Built by industry experts, our environments recreate long-horizon workflows we believe are critical to AGI. On our finance and coding benchmarks, top models pass just ~6% and ~52% of trials. Links in 🧵:

[Justin Jiang @jostinjiang]

@levensti @SherpaLabsAI @isaactruong @aaron_chang21 ty steven!

[Justin Jiang @jostinjiang]

@kevingu @SherpaLabsAI @isaactruong @aaron_chang21 luv u

[Justin Jiang @jostinjiang]

@latent_axioms @SherpaLabsAI 👀👀

[Justin Jiang @jostinjiang]

@carrynointerest @SherpaLabsAI @isaactruong @aaron_chang21 fixed mb

[Justin Jiang @jostinjiang]

@Selinaliyy 🙏🙏🙏

[Isaac Truong @isaactruong]

Thrilled to be buildling @SherpaLabsAI with @jostinjiang and @aaron_chang21. We build RL environments for long-horizon knowledge work, starting with finance, coding, and quant research.

Learn more about our benchmarks and what we're up to below in the 🧵:

[Aaron Chang @aaron_chang21]

Excited to be building @SherpaLabsAI with @jostinjiang and @isaactruong ! Today, we build RL environments for long-horizon knowledge work, developed with experts from the top firms in each industry, starting with finance, coding, and quant research.

On our finance and coding benchmarks, the strongest models achieve only a ~6% and ~52% pass rate. Learn more about us and our benchmarks below!

[Sherpa Labs @SherpaLabsAI]

Today, we are announcing Sherpa Labs. We build RL environments for knowledge work, starting with finance, coding, and quant research.

A real project runs for days, weeks, even months, spans dozens of interdependent steps and messy inputs, and relies on judgment that experienced practitioners pass down rather than publish. Our environments recreate this work as it happens in practice, with every one built by experienced practitioners from leading firms in their fields.

Our team brings experience from Surge AI, Millennium, Optiver, Blackstone, Qatalyst, Vista Equity Partners, and General Atlantic. We know what exceptional work looks like, and we hold our environments to that standard.

Backed by @ycombinator and others.

Read more about our benchmarks from the links in 🧵:

[Justin Jiang @jostinjiang]

@SherpaLabsAI @isaactruong @aaron_chang21 careers: https://t.co/eMugF6axdX

[Justin Jiang @jostinjiang]

@SherpaLabsAI @isaactruong @aaron_chang21 megafund-bench-1: https://t.co/boU8Pmzuru

[Justin Jiang @jostinjiang]

prod-bench-1: https://t.co/DKMxUauGY7

[Justin Jiang @jostinjiang]

@TychoSvoboda @joshkmartinez @hlumelonotshe @OpenAI @TensorPool congrats goats 🫡

[Tycho Svoboda @TychoSvoboda]

Another overdue announcement for your feed....@joshkmartinez , @hlumelonotshe and I have joined @OpenAI! We've had a blast building @TensorPool to make infra easy, and are pumped to keep pushing on that mission at an unprecedented scale.

HUGE thanks to our investors, users, and other folks that helped us along the way :)

[Kevin H. Zhao @ggkhzhao]

Thank you for waiting!

Our project is located here:
https://t.co/hF5bBZuYjU

What we used to create/run it:
- 4x NVIDIA L4s
- Codex[GPT5.5] + Claude Code[Opus4.8] (~$9k in credits)
- 2-3 weeks of our lives

See ".logbook" for a full history of experiments! Paper coming out later. We plan to also train neural correction on top just like NeuralGCM :)

We love open source and building on the shoulders of giants!