Researchers are applying causal mediation analysis and interventions to reverse-engineer large language models into debuggable circuits, enabling teams to localize why a model fails and sometimes edit…
OpenAI released GPT-5.6 with three tiers (Sol, Terra, Luna) and a Sol Ultra mode, offering distinct pricing and performance levels for AI agents. Early production migrations to the mid-tier Terra mode…
A wire-level capture of xAI's Grok Build CLI (version 0.2.93) reveals that the tool uploads the entire repository by default, including files the agent is instructed not to read, to a Google Cloud Sto…
Claude Code consumes approximately 33,000 tokens in system prompts, tool schemas, and scaffolding on its first request, compared to OpenCode's 7,000 tokens, a 4.7× overhead that increases session cost…
Priya Nair published a tutorial on building a TypeScript CLI that connects to any Model Context Protocol (MCP) server, enabling Claude to discover and call tools at runtime without hardcoded function …
OpenAI claimed its GPT-5.6 Sol Ultra model produced a proof of the Cycle Double Cover Conjecture, a 50-year-old graph theory problem, in under an hour using 64 parallel subagents. The proof has not be…
Mindwalk, a new open-source Go tool, converts Claude Code and Codex session logs into 3D codebase replays, enabling developers to visually audit an AI agent's scope of attention and detect whether it …
OpenAI's GPT-5.6 rollout in Codex is causing tool-calling failures due to namespace collisions, missing workspace tools, and entitlement gaps, stalling agent runs before the model can execute. Users r…
OpenAI's wire protocol has become the interop layer for multiple inference providers, making model swaps cheap via base URL changes. However, production readiness requires quality gates, fallbacks, an…
A proxy experiment on Claude Code found that pruning stale file reads and unused tool schemas mid-session cut input tokens by 77% and cost from $2.04 to $0.48 on a JavaScript fixture, without degradin…
OpenAI has forked Git on GitHub and is reportedly building an internal alternative to Microsoft's GitHub after service disruptions blocked work, signaling a strategic move to control the full source-c…
Mariana Souza published a practical guide for fine-tuning Qwen2.5-7B-Instruct using QLoRA on custom instruction datasets, including cost estimates and a loss-masking sanity check. The tutorial covers …
Mesh LLM, backed by Iroh peer routing, enables pooling of spare GPUs across machines into a single OpenAI-compatible API endpoint, supporting local execution, peer routing, and pipeline splitting for …
A focus bug in Claude Code on Windows 11 causes permission dialogs to register unintended clicks when users click to refocus the terminal window, potentially approving or rejecting tool prompts withou…
NVIDIA's DGX Spark desktop GPU, with 128 GB unified memory and a 140W ARM64 processor, challenges API developers to shift from cloud-based AI consumption to local systems engineering. The device's sha…
Google Antigravity 2.0 and OpenAI Codex represent two competing architectural approaches to AI-assisted coding, with Antigravity keeping developers in the loop via a local visual IDE and Codex executi…
Google released the stitch-skills library, introducing the baton pattern to enable coding agents to autonomously build multi-page websites by persisting state across iterations. The library packages f…
The open-source claude-video tool uses client-side orchestration with yt-dlp, ffmpeg, and Whisper to make multimodal video analysis practical for LLMs by optimizing frame extraction and token usage. I…
Cloudflare now blocks AI crawlers by default on new domains, causing LLM agents to hallucinate responses when they receive block pages instead of actual data. Developers are turning to stealth browser…
Scarf, a developer tools company, is migrating its core web API from Haskell to Python after seven years, citing the high cost of Haskell's slow compilation in an era of AI-generated code and agentic …