Twitxy Saved

Arshia Soltani

Verified account This account is verified. Learn more

@ckodser

MS&E PhD @ Stanford
Stanford, CA ckodser.github.io Joined August 2021
167 Following 128 Followers 65 Posts

Latest posts 5

Pinned post

Arshia Soltani Verified account @ckodser
We found a problem with on-policy self-distillation. The student makes mistakes early in a rollout, and the teacher has to continue from that flawed prefix, so its learning signal gets a lot weaker.

So we asked, "What if the teacher could step in before those mistakes happen?" Our method, Speculative Self-Distillation, lets the student generate while the teacher intervenes when their predictions diverge a lot.

We got better accuracy and training efficiency with our results, plus less forgetting in our tests. See @talaei_shayan’s thread for more!

Thanks to my amazing co-authors @chewing_a_gum, @jonashubotter, Amin Saberi and @Azaliamirh.
1:10
Shayan Talaei Verified account @talaei_shayan 5d
Introducing Speculative Self-Distillation (SSD, NeurIPS 2026), which pushes the accuracy–efficiency–forgetting frontier of self-distillation.

In SSD, student and teacher (the same model with additional information) write the rollouts together. Like a private tutor, the teacher lets the student write and takes over only when the student drifts too far. As a result, SSD trains on rollouts that stay as close to the student's own writing (on-policy) as possible while the teacher keeps them on track.

📉 38–69% fewer training tokens than on-policy, matching its accuracy
⏱️ Reaches on-policy accuracy in ~130 vs ~414 min (Science Q&A)
🧠 Forgets less in a stress test: −1.4 vs −4.5 points for on-policy
View quoted post on X
12 replies 15 reposts 155 likes 8.4K views View on X
Arshia Soltani Verified account @ckodser
Training a sota model with 5 people is crazy. I wonder if reaching this level needs <50 million and 5 researchers, does ANT really worth 1.5T?
Shayan Talaei Verified account @talaei_shayan Nov 4
Really happy to finally share Ripple-65B Preview! We trained it from scratch with a core team of five researchers, and it sets a new U.S. open-weight SOTA on Terminal-Bench 2.1 for agentic coding.

Three choices that shaped how we built it, and a few things I learned along the way.

1. We deliberately chose a dense model.

Many leading open-weight models are MoEs. For our use case, we wanted strong reasoning and low latency for agents going through long sequences of model and tool calls. Going dense also gave us a simpler training and serving stack and avoided expert-routing communication. That combination mattered a lot to us.

2. We bought no posttraining data.

A lot of the work went into building our own data generation and filtering pipelines, and making sure we covered the tasks we cared about. We built on open-source training and inference stacks throughout. I’m particularly happy about how much we were able to do as a small team by building on the community’s work.

3. We worked on the RL system and curriculum together.

Mixed RL meant dealing with very different rollout lengths, tool-call patterns and shares of the batch across domains. We tuned oversampling per domain and per problem to keep batch latency down, while managing policy staleness and trainer-inference mismatch. Score centering worked best among the corrections we tried; we used it alongside truncated importance sampling.

We also kept adjusting task selection and introducing harder problems as the model improved. Getting batches ready faster and keeping updates stable had to go hand in hand with giving the model useful problems to learn from.

Loved building this with @DZhang50, @esonghori2, @ethanboneh and @amiwilliamhuang. Huge thanks to our cofounders @annadgoldie and @Azaliamirh, and the whole @RicursiveAI family for making it possible. Really proud of what we built together.
View quoted post on X
2 replies 0 reposts 20 likes 1.4K views View on X
Arshia Soltani Verified account @ckodser
The author just confirms the bug our system detected in the STOC 2025 paper. AI reviews are getting real and even better than the STOC-level review process.
🚨We built an AI system that turns research-level math into Lean and machine-checks it.
We formalized 5 complex STOC papers to Lean, and even found a gap in one published proof.
Paper: arxiv.org/abs/2606.31134.
Thanks to @DARPA (expMath program) & @patrickshafto for their support.
View quoted post on X
0 replies 0 reposts 8 likes 1.2K views View on X
Arshia Soltani Verified account @ckodser
We built an automated pipeline that translates research math into formal languages. It successfully formalized 5 STOC papers! 🚀 It turns out AI systems are incredibly capable at translating complex human reasoning into machine-verifiable code.
🚨We built an AI system that turns research-level math into Lean and machine-checks it.
We formalized 5 complex STOC papers to Lean, and even found a gap in one published proof.
Paper: arxiv.org/abs/2606.31134.
Thanks to @DARPA (expMath program) & @patrickshafto for their support.
View quoted post on X
Arshia Soltani Verified account @ckodser
Excited to introduce Kaleidoscope, the largest benchmark for evaluating VLMs in culturally-authentic🌍, multilingual🗣️, and multimodal👀 settings!

Kaleidoscope includes 20,911 exams questions with:
🗣️ 18 languages
📚 14 subjects
📸 55% requiring image understanding!