Arshia Soltani (@ckodser) profile picture

MS&E PhD @ Stanford

Stanford, CA
Joined August 2021
Photo by Arshia Soltani (@ckodser)Photo by Arshia Soltani (@ckodser)Photo by Arshia Soltani (@ckodser)
Our Theo system formalized all 22 of Robertson and Seymour's Graph Minors papers, deepest theory in combinatorics: all 838 numbered results, no extra axioms. Including graph minor theorem (Wagner's conjecture), excluded-minor structure theorem, and O(n³) k-disjoint paths algo.
3
10
1
36
3,438
Arshia Soltani retweeted
On-policy learning is wasteful. Trajectories following an early mistake often don’t recover and waste many tokens on non-useful signal. We study speculative self-distillation: sampling as on-policy as possible while correcting “stupid” mistakes early. This improves the efficiency-accuracy-forgetting frontier. Every token counts! Awesome work with @talaei_shayan, @chewing_a_gum, @ckodser, Amin Saberi, and @Azaliamirh.
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
7
17
188
7,340
Arshia Soltani retweeted
Speculative Self-Distillation (SSD) lets AI models learn new knowledge from their own outputs, with a teacher stepping in only when predictions significantly diverge. The result: 38–69% fewer training tokens, near on-policy accuracy, and less catastrophic forgetting. Great work from @talaei_shayan 🚀
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
1
4
261
Arshia Soltani retweeted
New paper! On-policy self distillation is super wasteful. The teacher is only useful early in the rollout, and the student cannot course correct later. With SSD, we handhold the student, and see much faster convergence and lesser forgetting!
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
3
4
35
1,443
Arshia Soltani retweeted
Introducing Speculative Self-Distillation (SSD), which improves accuracy and effiecieny and reduces forgetting in self distillation. It turns out only a few tokens in a rollout actually depend on the new knowledge. Letting the teacher generate just those tokens results in nearly matching on-policy self-distillation with 38–69% fewer training tokens, and the model forgets less. Great work co-led by @talaei_shayan, @chewing_a_gum and @ckodser, with @jonashubotter and Amin Saberi (NeurIPS 2026). Check out Shayan’s thread for more.
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
8
33
2
311
12,348
Arshia Soltani retweeted
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
33
64
18
508
59,421
Today we release Ripple-65B Preview, our 65B dense LLM we pretrained from scratch. Ripple sets a new state-of-the-art for U.S. open-weight models in agentic coding tasks, surpassing significantly larger models including Muse Spark Open-816B, Beam-501B, Inkling-975B, and Nemotron 3 Ultra-550B. Learn more: ricursive.com/ripple
399
544
309
4,549
785,291
Arshia Soltani retweeted
Scaling self-verification with DeepSeek V4 Flash beats Claude Fable 5 on Terminal-Bench 2.1, while being 11x cheaper 💰 As open-source models become more capable, they can now generate large numbers of high-quality candidate solutions and verify their own outputs at very low cost. For example, we find that sampling just 5 solutions with DeepSeek V4 Flash and ranking them using the same model with LLM-as-a-Verifier can lead to a significant boost in accuracy (79% → 88%), outperforming closed frontier models on Terminal-Bench. Try it out today: github.com/llm-as-a-verifier… More on verification scaling in my previous post.
How can we extract richer signals from AI Feedback? Introducing LLM-as-a-Verifier✨— a simple verification scaling framework that achieves SOTA on agentic benchmarks 🚀 The key idea: - Use fine-grained scoring granularity (e.g., 1-20 instead of the standard 1-5 scale) - Take the expectation over the full logprob distribution of score tokens - Scale repeated evaluation and criteria decomposition You can use these fine-grained signals for more effective test-time scaling, RL, and agent monitoring! It achieves SOTA across Terminal-Bench V2, SWE-Bench Verified, RoboRewardBench, and MedAgentBench 👑 Advised by @Azaliamirh @istoica05 @drmapavone @chelseabfinn 🧵👇
155
399
119
3,114
1,034,897
Recently, OpenAI disproved planar unit distance problem, first posed by Erdős. We tested our pipeline (Theo) on it and Theo formalized the paper using under 15k lines of Lean, a big improvement over Aleph Prover's 33k lines & Open AI 1M lines. See beyondthelibrary.github.io/f…
5
5
59
8,840
🚨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.
1
6
4
42
9,423
🔊@UofMaryland computer science professor @MTHajiaghayi and Ph.D. students Arshia Soltani, Iman Gholami and Danny Mittal are developing AI systems that can generate, test and refine mathematical ideas with support from a $2.6M DARPA grant. Read more: go.umd.edu/Hajiaghayi-5-2026
2
2
8
5,966
Arshia Soltani retweeted
Our code for CoDE-Stop is out: github.com/sudoparsa/CoDE-St… Curious to see how it performs on your setup, feel free to reach out!
🚨 New Preprint! Large reasoning models rely on long chain-of-thought generation to solve complex problems, but this comes at a substantial compute cost. We introduce CoDE-Stop, a simple, training-free method that decides when to stop reasoning by tracking how model confidence evolves over time.
1
4
428
Arshia Soltani retweeted
🚨 New Preprint! Large reasoning models rely on long chain-of-thought generation to solve complex problems, but this comes at a substantial compute cost. We introduce CoDE-Stop, a simple, training-free method that decides when to stop reasoning by tracking how model confidence evolves over time.
1
6
2
28
6,173
Arshia Soltani retweeted
Excited to share that our paper, GHOST, has been accepted to #ICLR2026! GHOST connects diffusion models and multimodal LLMs to uncover hallucinations, using image generation to expose what models mistake for an object that is not actually there. Paper: arxiv.org/pdf/2509.25178
3
2
1
11
4,891
Arshia Soltani retweeted
🧐 Can language models explore hidden environments? (1/n) We introduce interactive tasks where models must probe for hidden info to find good solutions—but SOTA models often under-explore and end up sub-optimal. Demo: explore-exploit-bench.github… Paper: arxiv.org/abs/2601.22345
1
7
27
2,356
Putting Iran back onto the trajectory of a well functioning democracy would result in the greatest value creation ever.
I completely agree with Eren's assessment. I have many Iranian friends and few relatives who are among the smartest, highest-agency & nice people I know. Once the Iranian tyrants are deposed, I'm sure Iran will contribute greatly to humanity and prosper itself.
7
10
3
137
14,912
Arshia Soltani retweeted
«سقوط حکومت اسلامی به نفع ایالات متحده و جهان آزاد خواهد بود، و مردم ایران وضعیت بهتری خواهند داشت.» تا امروز، دوازده‌هزار ۱۲۰۰۰معترض توسط رژیم به قتل رسید‌ه‌اند آرش آرامش می‌گوید سقوط رژیم ایران به نفع مردم ایران و همچنین به سود امنیت جهانی خواهد بود
'The US will be better off, the free world will be better off, and the Iranian people will be better off.' Ari Aramesh says the fall of Iran’s regime would benefit both its people and global security.
377
2,308
83
5,817
224,640
The people of Iran are rising up! #freeiran
937
3,345
225
15,167
527,296
Arshia Soltani retweeted
A new trend has emerged among Iranian women.
779
6,581
1,497
52,501
7,121,962