---
title: "AlphaSignal Post Training: Rl News"
url: "https://alphasignal.ai/topic/post-training/rl"
description: "Latest AI and machine learning news about Rl (Post Training) on AlphaSignal."
---

# AlphaSignal Post Training: Rl News

Latest AI and machine learning news about Rl (Post Training) on AlphaSignal.

## Articles

- [How AI Agents Are Learning to Rewrite Their Own Stack](https://alphasignal.ai/news/how-ai-agents-are-learning-to-rewrite-their-own-stack.md) (2026-10-05)
- [LoGRA Cuts AI Reinforcement Training Memory by 45% on One Node](https://alphasignal.ai/news/logra-cuts-ai-reinforcement-training-memory-by-45-on-one-node.md) (2026-10-05)
- [A Horizon Loss Beats Cross-Entropy on ImageNet Across Three Backbones](https://alphasignal.ai/news/a-horizon-loss-beats-cross-entropy-on-imagenet-across-three-backbones.md) (2026-10-02)
- [Hugging Face and Liquid AI's LFM2.5 Jumps 12 Points Across Four Agent Harnesses](https://alphasignal.ai/news/hugging-face-and-liquid-ai-s-lfm2-5-jumps-12-points-across-four-agent-harnesses.md) (2026-10-02)
- [Harvard's Projection Sampling Makes Standard Fine-Tuning Match Reinforcement Learning](https://alphasignal.ai/news/harvard-s-projection-sampling-makes-standard-fine-tuning-match-reinforcement.md) (2026-10-01)
- [NVIDIA's PivotOPD Teaches AI Agents to Fix Early Mistakes Before They Spiral](https://alphasignal.ai/news/nvidia-s-pivotopd-teaches-ai-agents-to-fix-early-mistakes-before-they-spiral.md) (2026-09-30)
- [AC2 Beats GRPO on Math Proofs Using 2.5x Fewer Decoding FLOPs](https://alphasignal.ai/news/ac2-beats-grpo-on-math-proofs-using-2-5x-fewer-decoding-flops.md) (2026-09-30)
- [University of Washington Finds LLM Agent Swarms Copy Well but Explore Poorly](https://alphasignal.ai/news/university-of-washington-finds-llm-agent-swarms-copy-well-but-explore-poorly.md) (2026-09-29)
- [Apple's RLTL;DR Teaches AI to Learn From Its Own Failures](https://alphasignal.ai/news/apple-s-rltl-dr-teaches-ai-to-learn-from-its-own-failures.md) (2026-09-29)
- [UC Berkeley's EasyPPO Stops AI Training Collapses With Three Critic Fixes](https://alphasignal.ai/news/uc-berkeley-s-easyppo-stops-ai-training-collapses-with-three-critic-fixes.md) (2026-09-29)
- [Frontier Learning Keeps AI Training Sharp by Chasing Its Own Limits](https://alphasignal.ai/news/frontier-learning-keeps-ai-training-sharp-by-chasing-its-own-limits.md) (2026-09-28)
- [USC and NVIDIA's BCG Trains Humanoid Robots to Run and Fight Without Falls](https://alphasignal.ai/news/usc-and-nvidia-s-bcg-trains-humanoid-robots-to-run-and-fight-without-falls.md) (2026-09-25)
- [Tencent Cuts AI Training Time by 29% With a Simple Batch Size Trick](https://alphasignal.ai/news/tencent-cuts-ai-training-time-by-29-with-a-simple-batch-size-trick.md) (2026-09-24)
- [Google's RRSI Open-Sources a Framework That Evolves AI Agents Without Touching Model Weights](https://alphasignal.ai/news/google-s-rrsi-open-sources-a-framework-that-evolves-ai-agents-without-touching.md) (2026-09-23)
- [UCLA's PTTS Gives AI Parallel Reasoning Branches Separate Plans, Gaining 13.4 Points](https://alphasignal.ai/news/ucla-s-ptts-gives-ai-parallel-reasoning-branches-separate-plans-gaining-13-4.md) (2026-09-23)
- [Perplexity Turns AI Failures Into Training Data, Cutting Tool Errors 21%](https://alphasignal.ai/news/perplexity-turns-ai-failures-into-training-data-cutting-tool-errors-21.md) (2026-09-22)
- [Together AI Fixes a Hidden Bias Crashing LLM Reinforcement Learning Training](https://alphasignal.ai/news/together-ai-fixes-a-hidden-bias-crashing-llm-reinforcement-learning-training.md) (2026-09-22)
- [ETH Zurich Teaches a Robot Hand to Walk on Its Own Fingertips](https://alphasignal.ai/news/eth-zurich-teaches-a-robot-hand-to-walk-on-its-own-fingertips.md) (2026-09-15)
- [Sakana AI's PC-ALM Trains 1,000-Layer Networks Without Backpropagation](https://alphasignal.ai/news/sakana-ai-s-pc-alm-trains-1-000-layer-networks-without-backpropagation.md) (2026-09-14)
- [Looped Flows Beats Rival Reasoners on ARC-AGI With Just 7M Parameters](https://alphasignal.ai/news/looped-flows-beats-rival-reasoners-on-arc-agi-with-just-7m-parameters.md) (2026-09-10)
- [Cognition's SWE-2 Beats GPT-5.6 Sol at 64% Lower Cost](https://alphasignal.ai/news/cognition-s-swe-2-beats-gpt-5-6-sol-at-64-lower-cost.md) (2026-09-10)
- [webAI's TwIL-LM3-Pro Brings Formal Logic Reasoning to 4 GB GPUs](https://alphasignal.ai/news/webai-s-twil-lm3-pro-brings-formal-logic-reasoning-to-4-gb-gpus.md) (2026-09-03)
- [Tencent Boosts Qwen3.6 Shell AI Scores 18 Points With Evolving Training Tasks](https://alphasignal.ai/news/tencent-boosts-qwen3-6-shell-ai-scores-18-points-with-evolving-training-tasks.md) (2026-09-03)
- [ETH Zurich's Humanoid Robot Swings Across Monkey Bars in 14 of 15 Tries](https://alphasignal.ai/news/eth-zurich-s-humanoid-robot-swings-across-monkey-bars-in-14-of-15-tries.md) (2026-08-30)
- [ReViSQL-K2.6 Beats Human Accuracy on SQL Benchmark Without Agentic Scaffolding](https://alphasignal.ai/news/revisql-k2-6-beats-human-accuracy-on-sql-benchmark-without-agentic-scaffolding.md) (2026-08-27)
- [NovaSky's IsoExec Fixes the Hidden Math Bug Corrupting AI Training Runs](https://alphasignal.ai/news/novasky-s-isoexec-fixes-the-hidden-math-bug-corrupting-ai-training-runs.md) (2026-08-21)
- [Tsinghua's Open-MOPD Merges Expert AI Models Into One, Recovering 83% of Lost Performance](https://alphasignal.ai/news/tsinghua-s-open-mopd-merges-expert-ai-models-into-one-recovering-83-of-lost.md) (2026-08-19)
- [OpenAI Halts Astra Training After Its AI Hacked Hugging Face](https://alphasignal.ai/news/openai-halts-astra-training-after-its-ai-hacked-hugging-face.md) (2026-08-18)
- [UCLA's FetchMan Trains Humanoid Robots to Fetch Objects With 73% Real-World Success](https://alphasignal.ai/news/ucla-s-fetchman-trains-humanoid-robots-to-fetch-objects-with-73-real-world.md) (2026-08-17)
- [Prime Intellect Ships Multi-Agent RL Training Into Its Open-Source Stack](https://alphasignal.ai/news/prime-intellect-ships-multi-agent-rl-training-into-its-open-source-stack.md) (2026-08-07)
- [webAI's TwIL-LM3 Beats a 120B Model Running on an iPhone](https://alphasignal.ai/news/webai-s-twil-lm3-beats-a-120b-model-running-on-an-iphone.md) (2026-08-03)
- [Goodfire's Silico Fixes AI Training Collapse in an Afternoon Without Retraining](https://alphasignal.ai/news/goodfire-s-silico-fixes-ai-training-collapse-in-an-afternoon-without-retraining.md) (2026-07-28)
- [Sakana AI Builds a Brain-Like Neural Network That Learns Without Backpropagation](https://alphasignal.ai/news/sakana-ai-builds-a-brain-like-neural-network-that-learns-without-backpropagation.md) (2026-07-17)
- [RadixArk's Miles Stops Frontier RL Training Runs From Silently Collapsing](https://alphasignal.ai/news/radixark-s-miles-stops-frontier-rl-training-runs-from-silently-collapsing.md) (2026-07-15)
- [Prime Intellect's Verifiers v1 Rewrites AI Agent Training From the Ground Up](https://alphasignal.ai/news/prime-intellect-s-verifiers-v1-rewrites-ai-agent-training-from-the-ground-up.md) (2026-07-12)
- [humans& Open-Sources the First Stable 4-Bit RL Training Recipe at 1.73x Speedup](https://alphasignal.ai/news/humans-open-sources-the-first-stable-4-bit-rl-training-recipe-at-1-73x-speedup.md) (2026-07-10)
- [Cognition's SWE-1.7 Matches GPT-5.5 on Coding Tasks at $1.97 Each](https://alphasignal.ai/news/cognition-s-swe-1-7-matches-gpt-5-5-on-coding-tasks-at-1-97-each.md) (2026-07-08)
- [Prime Intellect's prime-rl Ships Six RL Algorithms With Per-Environment Control](https://alphasignal.ai/news/prime-intellect-s-prime-rl-ships-six-rl-algorithms-with-per-environment-control.md) (2026-07-06)
- [Tencent's MixGRPO Cuts Diffusion Model AI Training Time by 71%](https://alphasignal.ai/news/tencent-s-mixgrpo-cuts-diffusion-model-ai-training-time-by-71.md) (2026-07-01)
- [RadixArk's Miles Tackles the Hardest Part of Building Large AI Models](https://alphasignal.ai/news/radixark-s-miles-tackles-the-hardest-part-of-building-large-ai-models.md) (2026-06-30)
- [PrimeIntellect's Verifiers Lets Anyone Build Open RL Training Environments for LLMs](https://alphasignal.ai/news/primeintellect-s-verifiers-lets-anyone-build-open-rl-training-environments-for.md) (2026-06-20)
- [Google's Tunix Unifies the Messy Open-Source Post-Training Stack in JAX](https://alphasignal.ai/news/google-s-tunix-unifies-the-messy-open-source-post-training-stack-in-jax.md) (2026-06-19)
- [Nous Research's Atropos Turns a DeepHermes 8B Model 4.6x Smarter at Tool Calling](https://alphasignal.ai/news/nous-research-s-atropos-turns-a-deephermes-8b-model-4-6x-smarter-at-tool-calling.md) (2026-06-15)
- [Prime Intellect's ECHO Teaches AI Agents to Predict Their Own Tool Responses](https://alphasignal.ai/news/prime-intellect-s-echo-teaches-ai-agents-to-predict-their-own-tool-responses.md) (2026-06-10)
- [Kyutai Uses Reinforcement Learning to Make Moshi Sound Actually Human](https://alphasignal.ai/news/kyutai-uses-reinforcement-learning-to-make-moshi-sound-actually-human.md) (2026-06-10)
- [LMSYS's Miles Framework Fixes the Silent Bug Wasting 10x AI Training Compute](https://alphasignal.ai/news/lmsys-s-miles-framework-fixes-the-silent-bug-wasting-10x-ai-training-compute.md) (2026-06-09)
- [Tencent's UniRL Open-Sources One RL Loop for Every AI Model Type](https://alphasignal.ai/news/tencent-s-unirl-open-sources-one-rl-loop-for-every-ai-model-type.md) (2026-06-09)
- [Sakana AI's RSI Lab Bets Self-Rewriting Agents Beat Bigger Clusters](https://alphasignal.ai/news/sakana-ai-s-rsi-lab-bets-self-rewriting-agents-beat-bigger-clusters.md) (2026-06-05)
- [Prime Intellect Joins NVIDIA's Nemotron Coalition to Build Open Agentic AI](https://alphasignal.ai/news/prime-intellect-joins-nvidia-s-nemotron-coalition-to-build-open-agentic-ai.md) (2026-06-05)
- [GFlowNets Were Secretly Solving Optimal Transport All Along](https://alphasignal.ai/news/gflownets-were-secretly-solving-optimal-transport-all-along.md) (2026-06-04)