{"type":"post","id":"learning-new-skills-without-forgetting-self-distillation-vs-4031dd","slug":"learning-new-skills-without-forgetting-self-distillation-vs-4031dd","title":"Learning new skills without forgetting: Self-Distillation vs GRPO Deep Learning Engineering Gen AI LLM","description":"DoorDash compared self-distillation and GRPO for sequentially teaching Nemotron Nano 9B new skills while tracking performance on existing capabilities.","retrieval_nugget":"DoorDash compared self-distillation and GRPO for sequentially teaching Nemotron Nano 9B new skills while tracking performance on existing capabilities. In this experiment, GRPO learned the final logic task far better and preserved earlier benchmark performance more consistently, a reminder that continual post-training must be judged on retention as well as the new skill.","published_at":"2026-11-03","updated_at":"2026-11-03","record_date":"2026-11-03","date_kind":"published_at","topics":["skills"],"entities":["DoorDash","Chaitanya Ahuja","Sebastian Zuba","Nemotron Nano 9B","GRPO"],"related_patterns":["ai-engineering-evaluations-are-splitting-into-task-specific"],"novelty_label":"new_angle","what_changed":"Compares self-distillation with GRPO specifically for retaining previously learned capabilities during sequential skill training, finding GRPO preserves them better while self-distillation can cause forgetting and reasoning collapse.","prior_refs":["https://newruntime.com/posts/when-do-we-need-on-policy-distillation-distilling-ca2f7b/","https://newruntime.com/posts/jolt-extends-opsd-or-on-policy-self-distillation-7a545f/"],"source_url":"https://careersatdoordash.com/blog/learning-new-skills-without-forgetting-self-distillation-vs-grpo","source_urls":["https://careersatdoordash.com/blog/learning-new-skills-without-forgetting-self-distillation-vs-grpo"],"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:learning-new-skills-without-forgetting-self-distillation-vs-4031dd","status":"published","visuals":[],"editor":{"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]},"routes":{"html":"https://newruntime.com/posts/learning-new-skills-without-forgetting-self-distillation-vs-4031dd/","markdown":"https://newruntime.com/posts/learning-new-skills-without-forgetting-self-distillation-vs-4031dd.md","json":"https://newruntime.com/posts/learning-new-skills-without-forgetting-self-distillation-vs-4031dd.json"},"markdown_url":"https://newruntime.com/posts/learning-new-skills-without-forgetting-self-distillation-vs-4031dd.md","next_reads":[{"type":"topic","path":"/topics/skills/","url":"https://newruntime.com/topics/skills/","title":"Skills","reason":"Explore the skills topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/skills.md","json_url":"https://newruntime.com/topics/skills.json"},{"type":"collection","path":"/posts/","url":"https://newruntime.com/posts/","title":"Field Notes","reason":"Open the latest editorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/posts.json"},{"type":"collection","path":"/sources/","url":"https://newruntime.com/sources/","title":"Sources","reason":"See the original sources this archive cites.","media_type":"text/html"}]}