{"version":1,"type":"story","url":"https://digestai.news/story/perplexity-and-nvidia-unveil-jolt-a-joint-teacherstudent-rl-posttraini","json":"https://digestai.news/story/perplexity-and-nvidia-unveil-jolt-a-joint-teacherstudent-rl-posttraini.json","markdown":"https://digestai.news/story/perplexity-and-nvidia-unveil-jolt-a-joint-teacherstudent-rl-posttraini.md","slug":"perplexity-and-nvidia-unveil-jolt-a-joint-teacherstudent-rl-posttraini","headline":"Perplexity and NVIDIA unveil JOLT, a joint teacher‑student RL post‑training method","summary":"Perplexity and NVIDIA introduced JOLT, a post‑training method that lets a single model act as both teacher and student. The teacher receives privileged information while the student works only on the task, and both roles are trained simultaneously. JOLT+ adds the student’s own outcome reward to the update, aiming for a reward‑consistent teacher.\n\nIn benchmark tests across five settings—GSM8K, MATH, LiveCodeBench v6, AppWorld and Terminal‑Bench 2.1—JOLT+ achieved the highest mean scores, beating GRPO, OPSD and π‑Distill. On MATH, JOLT matched GRPO accuracy while using 6.5x fewer processed training tokens. On AppWorld, JOLT+ reached 72.09% versus GRPO’s 62.20%, a 9.9‑point gain. The method also showed gains of 1.5 points on GSM8K, 4.2 on MATH, 3.0 on LiveCodeBench v6 and 9.9 on AppWorld.","keyPoints":["JOLT matches GRPO accuracy on MATH while using 6.5x fewer processed training tokens.","JOLT+ reaches 72.09% on AppWorld, beating GRPO’s 62.20% by 9.9 points.","JOLT+ attains highest mean scores across GSM8K, MATH, LCBv6, AppWorld and Terminal‑Bench 2.1."],"whyItMatters":"By cutting compute needs up to 6.5x and improving benchmark scores, JOLT makes RL‑based post‑training of large language models cheaper and faster, lowering barriers for research labs and companies that fine‑tune models.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":["Perplexity","NVIDIA","Marktechpost AI Media Inc."],"models":["Qwen3-0.6B","Qwen3-4B-Instruct","Qwen3-8B","Qwen3-30B-A3B-Instruct","DeepSeek-V2-Lite"],"people":["Asif Razzaq"]},"firstPublishedAt":"2026-10-22T04:56:31Z","updatedAt":"2026-10-22T04:56:31Z","sourceCount":1,"hasPrimarySource":false,"sources":[{"outlet":"MarkTechPost","title":"Perplexity and NVIDIA Researchers Introduce JOLT: a Joint Teacher-Student Reinforcement Learning Method","url":"https://marktechpost.com/2026/10/21/perplexity-jolt-joint-on-policy-learning-teaching","publishedAt":"2026-10-22T04:56:31Z","type":"press","primary":false,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"Perplexity and NVIDIA unveil JOLT, a joint teacher‑student RL post‑training method\", 22 October 2026, https://digestai.news/story/perplexity-and-nvidia-unveil-jolt-a-joint-teacherstudent-rl-posttraini","publisher":"Digest AI","title":"Perplexity and NVIDIA unveil JOLT, a joint teacher‑student RL post‑training method","datePublished":"2026-10-22T04:56:31Z","url":"https://digestai.news/story/perplexity-and-nvidia-unveil-jolt-a-joint-teacherstudent-rl-posttraini"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}