269 items: records, experiences, guides and model families. Source reports remain distinct from discoveries.
discovery · Source published 2026-10-12
Wonder is available through Meta’s API and Muse Code. The release raises useful questions about persistent memory, long-running agents and how much a benchmark tells us about work we can trust.
discovery · Source published 2026-10-12
Software can decide when a home battery stores, uses or sells electricity. The useful question is what that changes for an eligible household, and what the evidence still cannot tell us.
discovery · Source published 2026-10-14
Two announced commitments offer compute access for UK safety research and technical support for Earthshot finalists. Delivery and resulting benefits still need evidence.
discovery · Source published 2026-10-14
New research shows how machine learning can improve the odds in three- and four-week weather outlooks by correcting forecasts that already exist.
discovery · Source published 2026-10-14
A Meta research sprint produced ten manuscripts in four weeks. Its more useful lesson may be how quickly researchers could reject weak ideas, and where expert checks still mattered.
discovery · Source published 2026-10-13
A small hospital study lets patients ask questions about their own brain recordings and help researchers improve how those signals are interpreted.
discovery · Source published date unknown
Faster replies can be useful. A fair test counts the whole customer problem, the work behind the tool and what actually changes in your costs.
discovery · Source published 2026-10-13
A free, reservation-only glasses service connects eight paintings with carefully prepared audio. The promise is easier access to art, with visitor benefits still to be measured.
discovery · Source published 2026-10-15
A statement of intent opens a new research effort. Existing particle-reconstruction work shows why the idea is promising, and how much evidence a broader model will need.
discovery · Source published 2026-10-15
The gated weight release gives developers another route to self-hosted AI. Memory requirements and commercial terms determine who can use it in practice.
discovery · Source published 2026-09-23
Klang’s released speech model offers a practical choice about where Swedish audio is processed. Its benchmark gains are promising, with important differences between speed and accuracy.
discovery · Source published 2026-10-15
A Caltech-led study pairs a model with memory and checks against physical constraints. Its cloud and airfoil tests offer a route toward earlier warnings, with real-world validation still ahead.
discovery · Source published 2026-10-15
Researchers used AI proposals and automated checks to improve circuit designs. The results concern estimated resource requirements, with trade-offs and verification limits that matter.
discovery · Source published 2026-10-18
OpenCode’s cofounder says his team needs agents working around the clock. The answer should be a faster buildout of accountable defense, not a choice between paralysis and blind trust.
discovery · Source published 2026-10-15
A practical follow-up for small teams running AI agents. Advisory published 15 October 2026; identified by Bright on 18 October 2026.
discovery · Source published 2026-10-19
Google DeepMind and Singapore’s GovTech are developing on-device scam protection. The goal is a timely warning while keeping conversations private. Public rollout is still ahead.
discovery · Source published 2026-10-19
The Gemma-based model lists Indonesian, English and four regional languages. Its public weights offer builders more choice, while language quality, computing needs and license terms still matter.
discovery · Source published 2026-10-19
A maintainer-confirmed repair shows what AI-assisted security can deliver. Anthropic’s October 19 update also raises a bigger question: how do we give open-source defenders more capability without simply giving them more work?
discovery · Source published 2026-10-19
New coding results invite a serious look at AI-assisted maintenance. Bright examines the evidence, the limits and what safer modernization would require.
discovery · Source published 2026-10-20
DeepSeek has opened DeepInfer, the engine behind V4.1-Flash. A plain-language guide to the release, its benchmarks, hardware needs and limits.