Full-text search across 70343 articles. Combine with topic and date filters; results sorted by relevance.
A new arXiv paper (2610.32060v1) presents a corpus-grounded taxonomy of communication failures in multi-agent LLM systems, built from 3,211 failed multi-round trajectories spanning medical QA, coding, finance, multi-hop …
A finite conformance study of tool-using AI agents found that retained-ledger reconciliation passed all 42 histories per host, while cached retry, revalidation before retry and stable-key retry passed 30, 36 and 31 histo…
A new arXiv paper (2610.31820v1) proves that confidence estimation and critic training in reinforcement learning are mathematically equivalent under a correctness-based outcome, meaning LLMs trained with value-based RL c…
A POS-signature demonstration-retrieval policy produced the highest labeled attachment scores among tested prompting systems for in-context dependency parsing, reaching 72.4 and 73.2 on the 2,077-sentence UD English Web …
Researchers released SOPO (Socratic Guided Policy Optimization), a multi-turn reinforcement learning framework for span-level hallucination detection in large language models, publishing training and evaluation code plus…
A new arXiv paper (2610.32492v1) introduces Accordion, a mixture-of-experts model that supports five inference budgets through fixed-prefix layer looping, with the minimum budget retaining the unlooped model. After 3.6T …
An open-source project called Freehire launched a searchable index of 3,712,951 unique tech jobs drawn from 536,962 companies across 370 sources and 156 ATS platforms, deduplicated and tagged by stack, seniority and loca…
Google's October 13, 2026 Gemini 4 Argon launch post claims the model was used to make Google's open Gemma 4 E2B nearly three times faster, achieving a 2.85x end-to-end inference speedup on an Nvidia A100 via vLLM that r…
An AI engineer outlined a practical methodology for stress-testing LLM-as-a-judge evaluation pipelines, warning that judges exhibit position bias, verbosity bias and reasoning errors that can make scores look like progre…
A developer released Muxer, an Android app that controls cmux terminal and agent sessions by routing HTTPS requests through a small relay to cmux's local socket. Muxer lets users answer agent permission requests, questio…
A developer documented an experiment testing how Claude Code, Codex, and Antigravity load instruction files, finding that Claude reads AGENTS.md only when CLAUDE.md contains an @AGENTS.md import line, that only Claude ex…
A GNOME user released three open-source projects built with help from Anthropic's Claude Code: howdy-fedora-helper-script, which sets up facial authentication on Fedora Workstation with GNOME from a single script; orrery…
A developer who builds and ships AI-generated apps in production published a free checklist of the failure modes that AI coding tools like Lovable, Bolt, Replit and v0 leave behind, including disabled Supabase row-level …
IQuest Research released IQuest-Q1, a 320B sparse mixture-of-experts open-weight model with a 512K-token context window and Claude Code integration. The release targets command-line agent workflows, with the company publ…
A developer released OpenIDEA, an open-source IDE built with Rust and Tauri that recreates IntelliJ-style workflows, including familiar keybindings, tool windows, and Search Everywhere, while aiming for fast startup and …
Moonshot AI has promised full weights for its Kimi K3.1 Pro model by Nov 4, 2026, but as of the announcement no repository, model card or license file exists in the company's Hugging Face organization, which holds 19 rep…
DirPlayer, a Rust-based Shockwave Player emulator, restores legacy browser games built on Shockwave .dcr files to modern platforms through a polyfill that detects and replaces <embed> and <object> elements, a Chrome exte…
Check Point Research published a report on October 23, 2026, detailing how the Aokigahara infostealer targets AI applications and developer tools including Claude, Codex and Cursor, collecting session data, configuration…
Arizona Vamp Recomp, a native PC port of the 2005 Xbox game Darkwatch built through static recompilation, has reached a beta state that is playable but not yet fully play-tested, with testing having reached the first cha…
A developer built Slothgram, an open-source tool that takes a single update and generates editable, per-destination versions for X, LinkedIn, DEV.to and GitHub, then publishes each as an independent job. The backend runs…