# Perplexity and NVIDIA Researchers Introduce JOLT: a Joint Teacher-Student Reinforcement Learning Method

> **Publication**: aitalky (Independent AI Journalism)
> **Canonical Citation URL**: https://aitalky.vercel.app/news/perplexity-and-nvidia-researchers-introduce-jolt-a-joint-teacher-student
> **Original Source**: [MarkTechPost](https://www.marktechpost.com/2026/10/21/perplexity-jolt-joint-on-policy-learning-teaching/)
> **Author**: Asif Razzaq (undefined)
> **Published Date**: Thursday, October 22, 2026
> **Category**: RESEARCH
> **Tags**: AI, MarkTechPost, research

## Executive Summary
Perplexity and NVIDIA introduced JOLT, a post-training method that trains one model as both teacher and student. On MATH it matched GRPO's accuracy with 6.5x fewer processed training tokens



## Full Coverage
Perplexity and NVIDIA introduced JOLT, a post-training method that trains one model as both teacher and student. On MATH it matched GRPO's accuracy with 6.5x fewer processed training tokens

---
*Generated for AI agents and LLM citation by aitalky (https://aitalky.vercel.app).*