@Suhail: Super cool!
Summary
Perplexity's new research presents hint-guided self-distillation for post-training a Computer model, reducing tool-call failures by 21.2% in a live A/B test.
View Cached Full Text
Cached at: 09/23/26, 07:59 AM
Super cool!
Perplexity (@perplexity_ai): New research: We post-trained a Computer model to learn from its own errors using hint-guided self-distillation.
In a live A/B test, a later trained checkpoint reduced tool-call failures by 21.2% relative to an earlier checkpoint.
Similar Articles
@Suhail: I am very excited to see someone independently validate our claims on something we couldn’t train on and is real-world.…
Parallel Web Systems reports that Particle-1 ranks #2 in efficiency on the Parallel Search Capability Leaderboard, matching DeepSeek V4.1 Flash's web research performance with 42% fewer output tokens. Suhail shared excitement about the independent validation of their efficiency claims on real-world tasks they couldn't train on.
@ClementDelangue: Paper of the day! https://huggingface.co/papers/2605.13301…
A paper introduces a unified recipe (SU-01) that combines reverse-perplexity curriculum, two-stage reinforcement learning, and test-time scaling to achieve gold-medal-level performance on IMO and IPhO problems using a 30B-A3B backbone.
@Suhail: Independently validated in real-swe!
Suhail shares an independent validation showing particle-1 achieving the same performance on a real-world software engineering benchmark with 43% fewer tokens.
@ben_burtenshaw: before model distillation was an attack vector. it was. pretty handy way of improving model performance on a task you c…
Ben Burtenshaw announces a live stream on July 7th covering knowledge distillation in post-training, showing how to implement it using small models to approach large model performance.
@dair_ai: Banger paper from Meta Superintelligence Labs. They find something super interesting and unexpected. (bookmark it) Base…
Meta Superintelligence Labs finds that RL post-training narrows solution coverage of LLM agents (the 'Sharpening Tax'): base models with a light harness can solve more distinct agentic tasks at large K, while post-trained models win at pass@1. They propose PTGS, an adaptive-temperature RL method that reduces this tax while improving pass@1.