Improving Small Language Model Reasoning With A* Search

ℹ️ EPISODE INFO

Released

August 6, 2026

Duration

6m

Channel

YC Root Access

Watch on YouTube

▶ Open ↗

📝 EPISODE DESCRIPTION

At our inaugural YCML at Startup School, YC Partner Ankit Gupta speaks with Alexander Braverman about a test-time scaling method for improving reasoning in smaller language models. Instead of relying on a larger teacher model or an external reward model, the method uses the language model’s own self-critique as a heuristic in an A*-inspired search. It explores multiple reasoning paths, deprioritizes weaker branches, and searches for a stronger answer using the same underlying model. On mathematical reasoning benchmarks, the method improved accuracy more efficiently than other test-time approaches at comparable token and runtime budgets. Apply to Y Combinator: https://www.ycombinator.com/apply Work at a startup: https://www.ycombinator.com/jobs

🎤 SPEAKERS

• Ankit Gupta

  host

  Host/interviewer on YC Root Access (Y Combinator).

• Alexander Braverman

  guest

  Researcher presenting work on improving small language model reasoning using A*-style test-time search and self-critique heuristics.

🧾 EPISODE SUMMARY

In this episode of YC Root Access, featuring Ankit Gupta and Alexander Braverman, Improving Small Language Model Reasoning With A* Search explores A* search uses self-critique to boost small-model reasoning accuracy The talk introduces a test-time scaling method that treats LLM reasoning as tree search and applies an A*-inspired algorithm to small language models.

🔗 RELATED EPISODES

Any-Horizon Reasoning for Video Agents

Diamond Maps: Efficient Reward Alignment for Generative Models

Interpretability and Safety for Robot Foundation Models

LeanAgent: Lifelong Learning for Formal Theorem Proving

The First Dedicated YC GPU Cluster - With Together AI

Building the Agentic AI Platform for Hospitals