Meta Built Forum in 8 Weeks by Letting an AI Agent Code
Meta's small Forum team let an AI agent own the plan-build-test-verify loop, reaching Facebook Groups feature parity in eight weeks.
20 stories
Meta's small Forum team let an AI agent own the plan-build-test-verify loop, reaching Facebook Groups feature parity in eight weeks.
A compact German named entity recognizer built on XLM-RoBERTa and document-level context hits 92.31 F1 on CoNLL-03 and crosses 948K downloads.
DeepMind's new vision-language-action model hits 96% accuracy on factory kitting and jumps 30-100% on dexterity tasks over its predecessor.
A standalone fork of Zed's GPUI framework reworks the layout engine to retain unchanged work between frames, cutting main-thread frame times by up to 55 percent.
An unofficial MCP server exposes 87 Home Assistant tools to Claude, ChatGPT, and other LLMs for building, debugging, and controlling smart homes.
A quantized GLM-5.3 variant trims 64 experts via neural architecture search, shrinking the model 24.6% so it fits on a single B200 GPU.
A new framework from Georgia Tech uses Bayesian uncertainty to tell SAM exactly where to ask for its next click, beating human annotators on 14 of 16 datasets.
Browser Use 1.0 scraps its DOM serializer and action schema, replacing them with a Python REPL running directly on Chrome DevTools Protocol.
A new evolution-strategy approach ditches random subspace projections for black-box prompt tuning and beats BBT across seven GLUE tasks.
An EPFL team gave frontier LLMs trained EEG and ECG network weights as text and asked them to debug, repair, and even write networks from scratch, without any gradient descent.
An ONNX port of BAAI's 33M-parameter embedding model runs entirely in the browser through Transformers.js, hitting 850K monthly downloads.
OpenAI's new Control Room service centralizes kill-switch authority over every research inference job and can halt model output within 35 seconds.
Anthropic's new frontier model lands at 174 on Epoch's Capabilities Index, five points above the prior leader and above the reasoning-model trendline.
ChatGPT dots can now be linked to several local machines at once, with per-task routing so a build can run on one Mac while you code on another.
Cursor adds Anthropic's updated Fable 5.2 model, pushing CursorBench to a new 60% ceiling while cutting token prices by 20% across the board.
Runway's new Sightline app generates a 360 panorama of a location, then lets you frame consistent shots from any angle and export for 3D use.
Epoch AI's new Inference Performance Index scores chips by memory and interconnect, not just FLOP/s, revealing GB300 serves 25x H100's load.
Anthropic's new frontier models formalized Fermat's Last Theorem in 30 hours, scored 49.1% on FrontierCode 2.0, and dropped prices 20%.
Meta's new efficient-tier model matches Claude Haiku 5.5 on intelligence at a third of the cost, with 1M context and native multimodal input.
World Labs' Atlas took the top spot on Rapidata's camera movement leaderboard, beating Google, Alibaba, and NVIDIA models across 513K human votes.