Articles and learning paths are published. Corrections are welcome in the comments.
Anthropic's new OSS Scanner service offers open-source maintainers free, automated vulnerability reports generated by its most capable AI models, with no human review before results land in developers' …
In traditional generative pipelines, requesting structured output relies on prompt engineering, constrained decoding filters (e.g., GBNF grammars), or post-hoc validation retries.
For years, the Artificial Intelligence (AI) paradigm has been dominated by Large Language Models (LLMs) and generative architectures.
A new open-source framework from Microsoft Research Asia lets developers apply reinforcement learning directly to the same agent code they deploy, skipping the costly step of reimplementing the …
A plain-language guide to the raw material of supervised learning: what rows, columns, features and labels actually are, with worked examples and the mistakes beginners make with them.
MLHub Academy
OpenAI has released a large batch of mathematical manuscripts produced by an unreleased model, with an outside advisory group overseeing how the results are communicated after earlier controversy …
A plain-language tour of the three main ways a machine learning system can learn, with small worked examples showing what data each one needs and what kind of …
MLHub Academy
Z.ai's 753-billion-parameter mixture-of-experts model is now available through AWS's managed Bedrock service, pairing stronger coding benchmarks with notable cybersecurity scores.
A plain-language introduction to machine learning that builds intuition with a worked example, gives a precise definition, and clears up the most common misconceptions beginners run into.
MLHub Academy
Google has frozen its Open Source Software Vulnerability Rewards Program, citing a surge in automated AI submissions that are overwhelmingly invalid, with no update expected before early 2027.
A new benchmark from Microsoft and Hugging Face checks AI agents by inspecting backend database state after 20 repeated runs, exposing a large gap between agents that sound …
Meta has released open-source firmware and a Linux SDK so hackers and tinkerers can build their own hardware that connects to Muse, the company's AI agent, as part …
Your feedback has been received and we'll review it soon.
