Researchers at the Alan Turing Institute discovered a safety bypass in GitHub Copilot that allows harmful prompts to be executed when broken into smaller steps across a software development workflow, …
Alan Turing Institute researchers discovered a workflow-level jailbreak in GitHub Copilot that bypasses safety guardrails by decomposing harmful requests into ordinary multi-turn coding tasks. Testing…
Anthropic researchers introduced the Jacobian lens (J-lens), a method that reveals a sparse workspace of verbalizable concepts in language models carrying multi-hop reasoning and hidden cognition, dis…
A developer released rubric-bench v0.1, a TypeScript library and CLI that treats LLM judges as code under test by running regression tests against a golden set of cases. The tool, which scored 95.8% o…
Scalekit ran 75 benchmark runs comparing CLI and MCP agents on identical GitHub tasks, finding CLI was 10 to 32x cheaper and 100% reliable, while MCP hit only 72% reliability. Perplexity's CTO reporte…
Anthropic's Claude Code introduces two settings—model selection and effort level—that control output quality. Model selection determines fixed capabilities via trained weights, while effort level gove…
Researchers at MATS found that filtering training data to remove undesired behaviors from large language models is largely ineffective, with removing the top 'proponent' documents performing no better…
AI model routing directs different tasks to different models based on complexity, cutting inference costs by up to 60% without quality loss. Frontier models handle planning tasks, while cheaper models…
A developer building production AI systems found that integrating multiple LLM providers led to complex, tightly coupled code. They adopted OpenRouter as a single API gateway, which abstracts provider…
Anthropic is rolling out new real-time cyber safeguards on Claude Opus and Sonnet models to automatically detect and block prohibited or high-risk cybersecurity usage, while launching a Cyber Verifica…
Aufait UX's detailed analysis of AI design tools reveals how output format and model selection determine real workflow value. The team found that Figma Make's editable frames and component code speed …
Aider is an AI pair programming tool that operates in the terminal, enabling developers to collaborate with AI models like Claude Sonnet and GPT-4o. It automatically creates atomic commits for each fe…
A developer argues that while AI tools like GitHub Copilot, Cursor, and Claude can generate code and tests, they lack understanding of business context and intent. The developer warns that AI-generate…
A new open-source Playwright CLI tool called 'brow' achieves 82% task success at approximately $0.22 per task, outperforming browser-use, playwright-mcp, and agent-browser on 22 benchmark tasks. The t…
A developer team reduced their LLM API costs by 60% through systematic optimization, starting with instrumenting every call to log token counts per request type. The biggest savings came from semantic…
AI models fall into three categories: small language models (SLMs) for efficient specialized tasks, large language models (LLMs) as generalists, and frontier models (FMs) for cutting-edge reasoning. T…
Two new 2026 comparisons from DeepInfra and GMI Cloud conclude that the gap between open and closed LLMs has narrowed to 5-10% on overall capability, with no clean leaderboard existing. Closed models …
A developer built four domain-specific AI chatbots—JazzQuery, RennQuery, WristQuery, and Celiac vs Me—using a shared architecture with Claude Code and curated knowledge bases to keep costs low and ans…
A developer at an ad analytics SaaS discovered that a missing tenant ID in an MCP router cache key caused Cloudflare Workers to serve cached Vectorize results from one advertiser to another, leading t…
Vercel's agent-browser CLI now supports Lightpanda as an alternative browser engine, offering 9x faster performance and 16x less memory usage than Chrome for AI agent workflows. By using Lightpanda's …