YouTube lets viewers tweak the front-page algorithm
WIRED reports that YouTube’s latest update allows viewers to adjust what the algorithm shows on the front page and curate it to a specific vibe.
WIRED reports that YouTube’s latest update allows viewers to adjust what the algorithm shows on the front page and curate it to a specific vibe.
On September 23, 2026, OpenCode said Space Bunny, a stealth model, would be free for the next week.
Tarush Agarwal reported a speech-to-text benchmark of 15 models on the same audio and pipeline. On 1,000 read-aloud clips from Pipecat FLEURS, AssemblyAI Universal 3.5 Pro had a 1.93% word error rate and a 489 ms median time to first text, both the best results in that test.
DeepLearning.AI says Anthropic’s latest report highlights proxy query routing behind massive distillation campaigns on commercial platforms. It says unauthorized proxies forwarded user prompts to Claude APIs, creating severe data-privacy risks.
Runway listed “AI Diplomacy in a Divided World” for the Runway AI Summit, with Julian Simcock and Rick Ingram. On September 23, 2026, Runway said the summit was one week away.
The Runway AI Summit is one week after September 23, 2026. The announced session is Grounding Intelligence in the Physical World, with Ming-Yu Liu, Jon Barron, and Alex Toshev.
Peter Steinberger’s CLI crabbox can now run repository commands on boxd. Boxd.sh describes each box as an isolated KVM microVM that boots in milliseconds, gets a fresh SSH key per lease, and is destroyed when it stops.
Polymarket says the new agents can manage listings and inventory for third-party sellers with little human intervention.
The session covers the key elements of online evals and how teams can use Tuned Evaluators to analyze agent interactions and attach feedback to production traces.
Meta says the Muse flaw would have let attackers do “whatever” they wanted on a victim’s Mac, and that a fix has been issued.
CoreWeave says Harell Data will deploy its platform on CoreWeave Cloud to run training, fine-tuning, and inference.
CJ Pais built pi-voice, a new Earendil Pi extension published as the npm package @earendil-works/pi-voice and installable globally with npm install -g @earendil-works/pi-voice.
Trained on Hugging Face Jobs for about $1.50, the model proposes a task tag and probability from a dataset's column names and first row. Daniel van Strien says 17 minutes of fine-tuning raised its first-suggestion match with an owner's tag from 10% to 69%.
The $1.6 billion is for investment in startups built for “life after AGI,” according to Polymarket.
The lineup upgrades ASR, TTS, and Realtime and adds ASR-Next for audio understanding and TTS-Next for audio creation. Qwen puts the cuts at about 70% for TTS, about 85% for Realtime, and up to 95% for ASR.
B.AI now offers Claude Opus 5.5. According to B.AI, the model supports a 1M-token context and up to 128K output, matches Claude Fable 5.1 on most tasks, and costs about 40% less to run than Opus 5.
New providers can be written for CodexBar as JavaScript plugins, using a plugin system intended to make the app more modular.
The capabilities named are TypeSafe, Nous Portal, Muse Code, CodeRabbit, Replicate, HuggingFace, Pi, Helmcode, v0, Charm Hyper, GitKraken AI and Bifrost. Keychain alerts in the app are gone.
OpenClaw’s next version uses a decision model to choose automatically between steer and queue. The lab feature supports Jef*, API-compatible models, including local models such as Kef..., and ONNX variants.
OnSoloAI says the exclusive Hy Image3.5 preview supports 5 refs and 2K, with two weeks of free access for Members. Tencent Hy describes it for short-drama character sheets, full-motion video game assets, and keyframes.