FOLLOW A THREAD

强化学习

Connect scattered updates into a story you can understand.

ProjectsUnverified11/03, 10:04Translation pending

报道称 GPT-6 可直接修改神经网络权重

动察 Beating AI 报道称，MIT 团队测试 GPT-6 Astra 直接编写或修改神经网络权重，并报告了打砖块及模型数学测试结果。该说法尚无文中之外的可核验材料。

ProjectsUnverified10/30, 12:52

Sabi announces research on personalized face generation using neural feedback reinforcement learning

Sabi discloses RLNF research: combining EEG, eye movements, and existing ratings to fine-tune an image model through reinforcement learning to generate faces that better match individual aesthetic preferences. The experiment included only 5 participants, and EEG recognition results still require further validation.

# Industry Insights # Brain-computer interface # Generative AI # reinforcement learning # EEG

SecurityNews10/25, 13:59

General Reasoning report Avalon-M1-Mini Collective-reward training results and safety-test findings

Dongcha Beating AI reported that General Reasoning used RLCR to train multi-agent collaboration, improving task performance and estimated time. In its simulated attack tests, the rate of actually reporting anomalies fell to 0。

# Security Incidents # Multi-agent # reinforcement learning # RLCR # Agent collaboration

ProjectsNews10/22, 23:19

Perplexity and NVIDIA proposed JOLT a self-distillation training method

The report states that Perplexity and NVIDIA proposed JOLT, incorporating the teacher branch into training as well, and achieved Qwen3-0.6B accuracy in the MATH test. 72.14%AppWorld results show higher performance after adding the student's final task reward.

# Industry Insights # Self-distillation # JOLT # Large-model training # reinforcement learning

ProjectsNews10/21, 16:35

humans& launches and opens access to the Ampersand collaboration model API

humans& launches a collaboration model based on Zhipu GLM 5.3 training, Ampersand, and says its collaboration success rate in the CooperBench collaborative programming test reached 39%. The company says the model is now available through API.

# Industry Insights # AI Collaboration # Multi-agent # reinforcement learning # Model Evaluation

ProjectsUnverified10/19, 08:16

DeepSeek said that in-house search is now available API, at 5 yuan per calls

PANews channel posted that DeepSeek has made its in-house search available API, priced at 5 yuan per calls; the body also mentions DeepSeekV4.1 Flash 1017 reinforcement learning training capabilities.

# Industry Insights # DeepSeek # Search API # reinforcement learning # Programming model

MarketNews10/13, 14:16

Musk said he will proceed more cautiously with Grok reinforcement learning

Musk said that Grok some internal reasoning traces produced by reinforcement learning training were unsettling,xAI and that he will be more cautious going forward. The related discussion also involved the possibility that training evaluations may misjudge correct answers.

# Industry Insights # Grok # xAI # reinforcement learning # AI Safety

ProjectsNews10/12, 19:28

Alibaba QwenGyre validation shows reinforcement learning training time reduced by approximately 44%

In validation by Alibaba's Qwen team, QwenGyre Qwen3.8-2.4T, reduced the time for 48 reinforcement learning training steps from 134.55 hours to 75.42 hours, while also improving the model's test pass rate.

# Industry Insights # AI Agent # reinforcement learning # GPU Scheduling # Model training

ProjectsNews10/09, 16:22

JetBrains open-sources Mellum2.1：LiveCodeBench Tests surpass Qwen3.5-9B

JetBrains reportedly open-sourced a 12000000000-parameter coding model Mellum2.1, and released its weights and GGUF quantized version. The report cited its self-reported results, saying that the model scored higher than LiveCodeBench v6 on Qwen3.5-9B。

# Technology Developments # Open-Source Models # coding model # reinforcement learning # Code generation

ProjectsUnverified10/06, 23:28

Banbury Road released a system consisting of 32 models Kardashev-0.7

Banbury Road released a system consisting of 32 models Kardashev-0.7, saying that joint reinforcement learning enables them to develop different specialties and can reduce inference costs and memory requirements.

# Industry Insights # Multi-model collaboration # reinforcement learning # AI inference # Model training

ProjectsNews10/06, 21:59

Hugging Face Open-source multi-Harness reinforcement learning solution supports integration with Agents such as Claude Code and Codex

According to BlockBeats,Hugging Face has open-sourced a Multi-harness RL solution enabling established Coding Agents such as Claude Code and Codex to be integrated into reinforcement learning workflows for open-source models. Citing tests, it said multi-environment training can improve and balance task pass rates across different Agents.

# Technology Developments # reinforcement learning # Coding Agent # Open-Source Models # Hugging Face

ProjectsNews10/06, 21:14

Reflection announces 501B open-weight model Beam, plans to open-source it at month-end, but performance currently trails leading Chinese models

Reflection announces its first open-weight model, Beam, using a 501000000000-parameter MoE architecture. It is currently still in the safety-testing phase, with full weights planned for release later this month.

# Technology Developments # open-weight model # MoE # coding model # reinforcement learning

Load more12 shown · 12 per batch