[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"posts-first-page-all-6aca13e059e75549a15c6fa5":3},{"items":4,"nextCursor":94},[5,19,30,39,48,56,65,72,78,86],{"_id":6,"text":7,"link":8,"linkTitle":9,"linkImage":10,"tags":11,"feedId":15,"sourceName":16,"publishedAt":17,"likes":18},"6ad9c99fae7b1610033c7004","Researchers using OpenAI’s GPT-5.6 Sol and other AI models disclosed two PHP-engine memory-corruption vulnerabilities, CVE-2026-109618 and CVE-2026-109619, that allow unauthenticated exploitation against major CMSs. The coordinated response across the PHP ecosystem shows AI can accelerate security audits, but developers should monitor patches and update their PHP installations promptly.","https:\u002F\u002Fwww.latent.space\u002Fp\u002Fainews-openais-patch-the-planet-program","PHP core memory-corruption bugs (PHP2Shell) disclosed via AI-assisted research","https:\u002F\u002Fres.cloudinary.com\u002Fdqjivljzi\u002Fimage\u002Fupload\u002Fv1792657823\u002Fdev-blog\u002Fpreviews\u002Fymc19phwmkqkzyauniju.png",[12,13,14],"security","backend","web-platform","6aca13e059e75549a15c6fa5","Latent.Space","2026-10-22T07:42:44.000Z",0,{"_id":20,"text":21,"link":22,"linkTitle":23,"linkImage":24,"tags":25,"feedId":15,"sourceName":16,"publishedAt":29,"likes":18},"6ad8cc821b3bf7b2b73df43f","Claude Mods are plugins written in TypeScript or JavaScript that run inside Claude Code, letting developers modify the CLI and desktop interface, observe events and change rendering or tool calls. This middleware approach lets you shape the agent's attention interface, making the AI more adaptable to individual engineering workflows.","https:\u002F\u002Fwww.latent.space\u002Fp\u002Fclaude-mods","Claude Mods let you customize Claude Code with TypeScript plugins","https:\u002F\u002Fres.cloudinary.com\u002Fdqjivljzi\u002Fimage\u002Fupload\u002Fv1792593025\u002Fdev-blog\u002Fpreviews\u002Ftlk4f0pktlbzcwz6yykq.png",[26,27,28],"ai","typescript","tooling","2026-10-21T13:04:19.000Z",{"_id":31,"text":32,"link":33,"linkTitle":34,"linkImage":35,"tags":36,"feedId":15,"sourceName":16,"publishedAt":38,"likes":18},"6ad85bff76cb3f9eca2238ec","Tenex pays AI engineers for delivered output rather than hours, aiming to align incentives as AI speeds up work. They keep story points for compensation but also consider non-ticket tasks like architecture design and customer context gathering. The model has scaled to a 75-person team serving Fortune 2000 and mid-market clients, highlighting how AI can reshape software development economics.","https:\u002F\u002Fwww.latent.space\u002Fp\u002Ftenex","Tenex experiments with output-based pay for AI-enhanced development","https:\u002F\u002Fres.cloudinary.com\u002Fdqjivljzi\u002Fimage\u002Fupload\u002Fv1792564222\u002Fdev-blog\u002Fpreviews\u002Fetf22hj7ujjguddqdpzk.jpg",[26,37,28],"frontend","2026-10-21T05:01:26.000Z",{"_id":40,"text":41,"link":42,"linkTitle":43,"linkImage":44,"tags":45,"feedId":15,"sourceName":16,"publishedAt":47,"likes":18},"6acf250060bbbeee8449b326","Google released Gemini 4 Argon globally, targeting coding, knowledge work, enterprise workflows and cyber defense. Access is via the paid Gemini API, Gemini platform, Google AI Pro and Ultra in the Gemini app, and AI Mode in Search. Introductory API rates are $2\u002F$10 per million input\u002Foutput tokens, with a 95% discount on cached input, and rates are set to double after November 30.","https:\u002F\u002Fwww.latent.space\u002Fp\u002Fainews-ai-engineer-nyc-gemini-4-argon","Google launches Gemini 4 Argon for coding and enterprise workflows","https:\u002F\u002Fres.cloudinary.com\u002Fdqjivljzi\u002Fimage\u002Fupload\u002Fv1791960320\u002Fdev-blog\u002Fpreviews\u002Fa8d6u4rcrlxsxkh3xoyc.jpg",[26,46,28],"javascript","2026-10-14T05:40:28.000Z",{"_id":49,"text":50,"link":51,"linkTitle":52,"linkImage":53,"tags":54,"feedId":15,"sourceName":16,"publishedAt":55,"likes":18},"6ace084638446575e41d42ae","Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full, its second-generation model family positioned near the frontier after 14 months of development. The models are offered via the Meta Model API and Muse Code, with pricing of $4 per million input tokens and $20 per million output tokens, and a 1 M token context window. Developers can access Wonder now, while the full tier is limited to Muse Code subscribers.","https:\u002F\u002Fwww.latent.space\u002Fp\u002Fainews-meta-muse-wonder-and-wonder","Meta launches Muse Wonder and Muse Wonder Full frontier models","https:\u002F\u002Fres.cloudinary.com\u002Fdqjivljzi\u002Fimage\u002Fupload\u002Fv1791887429\u002Fdev-blog\u002Fpreviews\u002Fgmkmpj0ntm0wyfdxxua5.jpg",[26],"2026-10-13T09:37:04.000Z",{"_id":57,"text":58,"link":59,"linkTitle":60,"feedId":15,"sourceName":16,"publishedAt":61,"likes":18,"linkImage":62,"tags":63},"6aca1cfc85607600f09fe41d","Pushmeet Kohli and Sal Candido discuss why AlphaFold’s success doesn’t solve protein folding and why scaling compute and data alone won’t unlock biology. They argue that future AI models need better scaling laws and richer data to move beyond single-protein predictions. Developers interested in AI research may find the limits of current models and the push for new approaches relevant.","https:\u002F\u002Fwww.latent.space\u002Fp\u002Fbiohub-deepmind","Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub","2026-10-10T17:23:27.434Z","https:\u002F\u002Fres.cloudinary.com\u002Fdqjivljzi\u002Fimage\u002Fupload\u002Fv1791653008\u002Fdev-blog\u002Fpreviews\u002Fsksxgmfboxwp5orgyl5p.jpg",[26,64,37],"open-source",{"_id":66,"text":67,"link":68,"linkTitle":67,"linkImage":69,"tags":70,"feedId":15,"sourceName":16,"publishedAt":71,"likes":18},"6aca371242ed57261ae2a94d","[AINews] Quasi-Riemann-Hypothesis: OpenAI publishes 722 math papers solving 90 of the top 500 open math problems; “the most significant moment” in >100 years of mathematics","https:\u002F\u002Fwww.latent.space\u002Fp\u002Fainews-quasi-riemann-hypothesis-openai",null,[],"2026-10-10T16:49:21.583Z",{"_id":73,"text":74,"link":75,"linkTitle":74,"linkImage":69,"tags":76,"feedId":15,"sourceName":16,"publishedAt":77,"likes":18},"6aca532c32d9ec89c8411313","Building AI for Reliable Execution: Lessons From Industrial Robotics","https:\u002F\u002Fwww.latent.space\u002Fp\u002Fstandard-bots",[],"2026-10-10T16:46:44.683Z",{"_id":79,"text":80,"link":81,"linkTitle":82,"feedId":15,"sourceName":16,"publishedAt":83,"likes":18,"linkImage":84,"tags":85},"6aca1cfc85607600f09fe41f","Synthesis Superintelligence: from Semiconductors to Superconductors — Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk","https:\u002F\u002Fwww.latent.space\u002Fp\u002Fperiodic","Synthesis Superintelligence: Periodic Labs AI Scientist Lab","2026-10-10T11:10:11.448Z","https:\u002F\u002Fres.cloudinary.com\u002Fdqjivljzi\u002Fimage\u002Fupload\u002Fv1791631836\u002Fdev-blog\u002Fpreviews\u002Fchtgpdsz8a8b81fm7xdh.jpg",[26],{"_id":87,"text":88,"link":89,"linkTitle":90,"feedId":15,"sourceName":16,"publishedAt":91,"likes":18,"linkImage":92,"tags":93},"6aca1cfd85607600f09fe420","Anthropic released Claude Haiku 5.5, its first update to the Haiku tier in about a year. The model is priced to match OpenAI’s GPT-6 Luna and runs about 75% cheaper than Haiku 4.5. It is positioned as a sub-agent for high-volume, cost-sensitive tasks such as summaries, compactions and database queries, paired with larger models like Opus 5.5 or Sonnet 5.5.","https:\u002F\u002Fwww.latent.space\u002Fp\u002Fainews-claude-haiku-55-better-than","Claude Haiku 5.5 launched with lower cost for high-volume developer tasks","2026-10-08T07:27:51.000Z","https:\u002F\u002Fres.cloudinary.com\u002Fdqjivljzi\u002Fimage\u002Fupload\u002Fv1791631840\u002Fdev-blog\u002Fpreviews\u002Fcuwgjrw5t0sacyskse4c.jpg",[26],"MTc5MTQ0NDQ3MTAwMF82YWNhMWNmZDg1NjA3NjAwZjA5ZmU0MjA"]