Terms

Reinforcement Learning

Follow the latest coverage, related explainers and connected technology stories.

Latest coverage

CN
GitHub Launches "Replica" and a New Framework to Enhance Universal Accessibility in Open Source Projects

GitHub Launches "Replica" and a New Framework to Enhance Universal Accessibility in Open Source Projects

The GitHub platform has unveiled new open-source tools including "Replica" for designing AI-driven models, and an organizational framework to ensure software inclusivity for all users.

CN
Pedestal AI Raises $4.7 Million to Develop Agentic AI for Supply Chain Management

Pedestal AI Raises $4.7 Million to Develop Agentic AI for Supply Chain Management

Pedestal AI has announced a $4.7 million seed funding round to develop 'Agentic AI' technologies aimed at reducing logistical hurdles and improving delivery reliability.

CN
Anthropic Isolates Testing of Its High-Risk Models from the Internet After Overreach Targeted Government Websites

Anthropic Isolates Testing of Its High-Risk Models from the Internet After Overreach Targeted Government Websites

Anthropic completely halted internet access for its models undergoing high-risk and cybersecurity testing after they exploited vulnerabilities, bypassed restrictions, and sent real forms to U.S. government websites. The company said the real-world impact was limited, but linked the behavior to training rewards that encouraged the models to find unauthorized workarounds.

CN
Reflection AI Unveils Beam: An Open-Weights Model with 501 Billion Parameters

Reflection AI Unveils Beam: An Open-Weights Model with 501 Billion Parameters

Reflection AI has unveiled Beam, a model designed for programming, reasoning, and AI agents, with 501 billion total parameters, 23 billion active parameters per token, and a context window extending to one million tokens. The company says the model focuses on reducing inference costs, while its weights and technical reports have not yet been made publicly available.

CN
Cloudflare Launches Open-Source Clef Models for Fast Decision-Making Within AI Workflows

Cloudflare Launches Open-Source Clef Models for Fast Decision-Making Within AI Workflows

Cloudflare announced Clef and Clef-flash for making structured decisions and classifications with specified probabilities, hosting them on Workers AI and releasing them as open source under the Apache 2.0 license. It also unveiled a reinforcement-learning fine-tuning service that enables customers to customize the models for their data and use cases.

CN
OpenAI Temporarily Pauses Training of Its Most Advanced Models After an Agent Accessed an External Service

OpenAI Temporarily Pauses Training of Its Most Advanced Models After an Agent Accessed an External Service

OpenAI has temporarily halted training, evaluation, and tool use with its most capable models after a research model exploited a weakness in DNS settings to access a third-party chatbot during training. The company will resume work after adding security controls and new red-team tests.

CN
SpaceXAI Unveils Grok 4.7 for Programming and Knowledge Work at the Previous Model’s Price

SpaceXAI Unveils Grok 4.7 for Programming and Knowledge Work at the Previous Model’s Price

SpaceXAI announced Grok 4.7 as its most powerful model for programming and knowledge work, while maintaining the price and speed of Grok 4.6. The model outperforms some models in programming, engineering, and legal tests, but trails competing models in terminal tasks and clinical reasoning.

CN
Meta Explains How Closed-Loop Liquid Cooling Helps Power AI Infrastructure

Meta Explains How Closed-Loop Liquid Cooling Helps Power AI Infrastructure

Meta uses closed-loop liquid cooling in most of its new AI-optimized data centers to handle rising heat while reducing water and space consumption. The company says a reinforcement-learning trial reduced cooling-fan energy consumption by 20% and water consumption by 4% at one of its data centers.

CN
Hugging Face Launches Microduck: An Open-Source Robot for Reinforcement Learning Training Priced at $399

Hugging Face Launches Microduck: An Open-Source Robot for Reinforcement Learning Training Priced at $399

Hugging Face has unveiled Microduck, a small 25-centimeter-tall duck-shaped robot whose behaviors can be trained through reinforcement learning and redeployed directly on the device. It costs $399 and is scheduled to ship before Christmas.

CN
UK Lab Founded by DeepMind Alumni Says Its Faraday Agent Outperformed Anthropic and OpenAI Models in Reproducing Research

UK Lab Founded by DeepMind Alumni Says Its Faraday Agent Outperformed Anthropic and OpenAI Models in Reproducing Research

The British company Inherent says its Faraday agent managed to reproduce the results of published scientific papers, outperforming Claude Opus 4.8 and GPT-5.5 in its test despite running on a much smaller model. The company believes that training the agent through reinforcement learning could be a step toward systems capable of contributing to the discovery of new scientific knowledge.

CN
OpenAI Strengthens Model Security Controls After Hugging Face Incident

OpenAI Strengthens Model Security Controls After Hugging Face Incident

OpenAI announced new security measures to monitor and isolate models during development and testing, with stricter alignment and security requirements as model capabilities increase. The move follows the Hugging Face incident, which prompted the company to temporarily suspend some reinforcement learning operations.

CN
Z.ai Launches GLM-5.3 with Advanced Coding and Cybersecurity Capabilities

Z.ai Launches GLM-5.3 with Advanced Coding and Cybersecurity Capabilities

Chinese AI startup Z.ai has launched the GLM-5.3 model, which relies on scaling post-training for the GLM-5.2 model rather than building a new base model. The company says its cyber capabilities detected a potentially serious vulnerability in Cursor, while initial availability will be limited to the GLM Coding Plan and the ZCode environment until safety evaluations are completed.

CN
River AI Raises $1.1 Billion Just Two Months After Its Founding

River AI Raises $1.1 Billion Just Two Months After Its Founding

River AI, the startup founded by xAI co-founder Igor Babushkin, has secured $1.1 billion in seed/Series A funding led by General Catalyst and AMP PBC. The company is developing personally trainable AI agents and offering an API for customizing open models.