TrainLoop

Post training research

[TrainLoop @TrainLoop_ai]

@nollahealth is pioneering access to care with AI-driven prescriptions. But critical use cases like these require superhuman performance.

Together with @nvidia and @nebius, we post-trained the new Nemotron 3.5 VL Super to SoTA on differential diagnosis.

[Jackson Stokes @jackson_stokes]

We partnered with @nvidia, @nollahealth, and @nebiusai to post train Nemotron 3.5 VL Super far past the frontier on dermatology diagnosis.

The model set new standards in our internal measures of trainability, and with Nolla it's powering the future of AI in medical care.

[Nolla Health @nollahealth]

2,500+ sign-ups, 1,200+ attendees, and so much love! 💙 Our Dermalogica x Nolla pop-up took over the West Village with skin scans, skincare, and more!

Huge thanks to everyone who came out and @dermalogica for making it happen. More to come soon.

[TrainLoop @TrainLoop_ai]

TrainLoop is proud to sign Open Weights and American AI Leadership.

As our economy transitions to the Autonomy Age, the best organizations will chart their own path, building their intelligence foundation atop open models.

https://t.co/Se2mZHhWF1

[Jackson Stokes @jackson_stokes]

We partnered with @mercor to test a simple idea:
What if knowledge-work agents were just… coding agents?

Result: +25% performance, 2x faster, cheaper, and new SOTA on APEX-Agents. @josancamon19

[TrainLoop @TrainLoop_ai]

+48% from a single RL step, and 100k rollouts from a single policy.
@LoganGrasby wrote up his findings that off policy training with OAPL is more robust than we thought!

[Jackson Stokes @jackson_stokes]

can we train a model in single RL step?

During recent experiments, @Logangrasby found that a single step of OAPL increased model performance from ~0 to 48% on a clinical reasoning and prediction task.
Turns out, data staleness might matter less than we think. with @pathos :

[TrainLoop @TrainLoop_ai]

Task-specific training can be far more efficient than we realize. This work by @hasith_v explores the LoRA-trained solution space for GSM8k, finding a massive "plane" of solutions, representable in a single rank.

[Jackson Stokes @jackson_stokes]

We trained LoRA adapters of different ranks to understand training dynamics, finding that adapters for GSM8k live in a surprisingly vast, low-rank solution space.

This hints that some model skills are easy to learn, and training is more forgiving than we think. @hasith_v 1/6 🧵

[TrainLoop @TrainLoop_ai]

Specialized models for the most important tasks!

[Jackson Stokes @jackson_stokes]

We post-trained MedGemma to be SoTA in visual medicine ddx, outperforming Opus 4.6, Gemini 3.1 and GPT-5.4 while running at ~1/30th the cost. @nollahealth Part 1 - improving visual reasoning 🧵1/6

[Joan Cabezas @josancamon19]

🧵 Labs and VC's are throwing cash at RL environments, especially for computer and browser use. Yet, with just 4 customers and over 30+ vendors, is cloning every website in the world really the path to scale? of course not.

Introducing TRACE: Trajectory Recording and Capture of Environments.

[TrainLoop @TrainLoop_ai]

@FlyaKiet @jackson_stokes @mlpierce22 @avimakesrobots @superset_sh TrainLoop ❤️🤝 Superset

[TrainLoop @TrainLoop_ai]

@jackson_stokes @gdb No way

[TrainLoop @TrainLoop_ai]

@jackson_stokes chatgpt is merging with windsurf is merging with vim

[TrainLoop @TrainLoop_ai]

@elonmusk @EnhancedRadar

[TrainLoop @TrainLoop_ai]

@ThePrimeagen hold my beer