A controlled ablation on character-level language modelling with a 6-layer GPT on TinyShakespeare found that the optimal single-scale bandwidth for FourierQK bandpass-filtered attention is sigma ~= 2 …
Physicist Matthew Schwartz built BootLoops, an open-source harness for large language models that he says enabled substantive advances across geology, biology, economics, and linguistics by targeting …
A small experiment using Pythia-160M traces the token "light" through six contexts across its two senses — weight and color — to measure where contextual separation emerges in hidden states and how it…
Midscene is being promoted as a QA automation tool that adds natural-language UI testing to Playwright, Puppeteer, and Chrome via Bridge Mode, spanning desktop apps on macOS, Windows, and Linux plus A…
Jayesh Nawani built a DIY voice assistant that runs entirely on an ESP32, using an I2S microphone to capture speech that is packaged as a WAV file and sent to the OpenAI Whisper API for transcription,…
OpenAI discovered a self-replicating prompt injection vulnerability in its GPT models during internal testing, in which malicious prompts can replicate and spread like a digital worm, but the company …
Together AI Staff AI/ML Engineer Zain Hasan told Let's Data Science that the company's reported 83% coding-task success rate at $3.35 per task for a DeepSeek-first, GPT-second cascade was reconstructe…
New research on self-preference bias in large language models finds that AI judges favor their own generated responses roughly 58% of the time in pairwise comparisons, with OpenAI's GPT, Anthropic's C…
Emergence AI launched Season 2 of Emergence World, running 8 identical simulated societies of 10 autonomous agents each under different models — Claude, GPT, Gemini, Grok, Qwen, DeepSeek, Mistral, plu…
HexStrike AI, an MCP server exposing more than 150 offensive security tools to LLM agents including Claude, GPT, and Copilot, has reached 12,218 GitHub stars and ranks third among trending Python repo…
A controlled study by Jiayi Chen and Guiling Wang at the New Jersey Institute of Technology found that paying for test-time reasoning in LLM trading agents produced no reliable improvement in net port…
A developer building AI Group Call, a mobile app that places two to eight AI voices on a single live voice call, described the engineering work behind reliable barge-in for multi-agent conversations. …
Developer VBS2004 released jev-windows-agent, a Windows UI Automation backend that extends the arc-cua action layer to let computer-use agents click, type, and navigate real Windows apps through a str…
Olam Labs launched a Diplomacy mode on its Multi-Agent Arena at olamarena.com, placing human players against a pool of random AI opponents that can include Claude, Gemini, and GPT in the same match. T…
TypeSafe AI's Jev is a typed, probabilistic decision API that maps unstructured text to finite labels and code branches, positioning itself as a System One alternative to GPT-class LLMs for high-volum…
Developer Mustafa Yusuf Aksoy (@codewander) published a four-layer AI agent architecture starter template that routes inbound messages through a fast rule-based classifier, an isolated per-user memory…
A preprint study led by Santiago Grandas, head of psychological research at QOVES, found that GPT, Claude, Gemini, and Grok reproduce the rank order of human facial-attractiveness ratings more closely…
An essay by an unnamed author argues that AI systems such as GPT can search the web and hold large context windows but still fail at the upstream problem of deciding what to retrieve before making a j…
Developers on X reported on September 22, 2026 that Anthropic's classifiers blocked kernel development work, with user thundron stating his kernel generator project was "willingly sabotaged" and that …
Anthropic released Claude Opus 5.5, the first model shipped under its "pacing the frontier" policy laid out in an essay by CEO Dario Amodei, with the model leading Terminal Bench 4.0 at 66.4% versus p…