LLMs can invent their own compression - Rajan Agarwal

kept by eddie nov 26, 25

Co-training a summarizer and generator via RL lets an LLM learn its own text compression scheme, achieving ~9.5% context size with minimal prediction loss.

joke-generator
Training a joke generator on Kimi K2 using rubric-based RL that decomposes humor into verifiable properties like specificity and commitment.
also on Tinker, #reinforcement-learning, #machine-learning, #ai-research
Learning to Replicate Expert Judgment in Financial Tasks - Thinking Machines Lab
Thinking Machines Lab fine-tuned a custom model on expert-labeled financial data to outperform frontier LLMs on information-filtering tasks at lower cost.
also on Tinker, #machine-learning
AI infrastructure in the "Era of experience"
As open-source LLMs commoditize due to fierce Chinese competition, companies can build defensible moats through custom RL models trained on proprietary environments using LoRA and GRPO.
also on #reinforcement-learning, Qwen, #ai-research
GitHub - THUDM/CaRR: This repository contains the code and data for the paper "Chaining the Evidence: Robust Reinforcement Learning for Deep Search Agents with Citation-Aware Rubric Rewards".
Citation-aware rubric rewards (CaRR) and C-GRPO training improve deep search agents' reasoning quality and factual grounding over standard binary outcome rewards.
also on #reinforcement-learning, Qwen
9 theses on AI | Sarthak Munshi
AI progress is constrained by long-task reliability, labor reallocation, cost inefficiencies of general APIs, the declining value of raw coding skills, inadequate benchmark testing, the limits of formal verification without strong specs, memory-bound local hardware advantages, the shift from data to environment-driven training, and the rising competitiveness of US open-weight models.
also on Qwen, #machine-learning