Loading the page.
./news · auto-feed
updated 07:15 PM MSK
● live
AI and neural network news
The machine selects the important stuff from the stream: releases, prices, tips. Nothing extra. RSS
From an articlewired.com10/08
Adjust calculations with sliders in ChatGPT: Intelligent UI creates interactive answers
From an articlewired.com10/08Adjust calculations with sliders in ChatGPT: Intelligent UI creates interactive answers
From an articlegithub.com10/08
Claude Code stops confusing prohibitions with permissions: instruction hooks fixed
From an articlegithub.com10/08
Claude Code stops confusing prohibitions with permissions: instruction hooks fixed
From a tweet@addyosmani10/08
Subagent tasks get cheaper: Haiku 5.5 costs 75% less than Haiku 4.5 on average
«Claude Haiku 5.5 is here 🎉 It's an affordable, fast and compact model with a broad range of capabilities. It costs roughly 75% less to use than Haiku 4.5, and you can adjust the model's effort. Our team really enjoys using it as a subagent alongside Opus 5.5.»
Addy Osmani
From a tweet@addyosmani10/08Subagent tasks get cheaper: Haiku 5.5 costs 75% less than Haiku 4.5 on average
«Claude Haiku 5.5 is here 🎉 It's an affordable, fast and compact model with a broad range of capabilities. It costs roughly 75% less to use than Haiku 4.5, and you can adjust the model's effort. Our team really enjoys using it as a subagent alongside Opus 5.5.»
Addy Osmani
From an articlegithub.com10/08
Gemini CLI login without endless retries: nightly build fixes OAuth loops
▶
From a tweet@artificialanlys10/08
Artificial Analysis adds a filter to compare base video models only
«Derived models are now labeled in the Artificial Analysis video model rankings, and you can hide them. Our rankings include a growing numb…» — Artificial Analysis
From a tweet@artificialanlys10/08Artificial Analysis adds a filter to compare base video models only
▶«Derived models are now labeled in the Artificial Analysis video model rankings, and you can hide them. Our rankings include a growing numb…» — Artificial Analysis
▶
From a tweet@artificialanlys10/08
Vidu Q4 Preview breaks into the top three for image-to-video generation: pricing stays the same
«Vidu Q4 Preview debuts at #3 in the Artificial Analysis AA-Video-I2V v1.0 rankings: a jump from Vidu Q3 Pro’s #19 spot. Vidu Q4 Preview is…» — Artificial Analysis
From a tweet@artificialanlys10/08Vidu Q4 Preview breaks into the top three for image-to-video generation: pricing stays the same
▶«Vidu Q4 Preview debuts at #3 in the Artificial Analysis AA-Video-I2V v1.0 rankings: a jump from Vidu Q3 Pro’s #19 spot. Vidu Q4 Preview is…» — Artificial Analysis
From a tweet@_catwu10/08
Claude finds users to interview: the week's 10 most active
«One of my favorite ways to use Claude as a product manager is to ask: "Who used <feature> most often last week? Make me an artifact with th…» — cat
From a tweet@_catwu10/08Claude finds users to interview: the week's 10 most active
«One of my favorite ways to use Claude as a product manager is to ask: "Who used <feature> most often last week? Make me an artifact with th…» — cat
From a tweet@rauchg10/08
AI agents need a stopping point: even unnecessary code can be improved forever
«The engineering community will have to rediscover one thing: almost any program can be improved practically forever by ① making it more resilient to failures (fixing edge-case bugs, validating inputs more strictly, handling errors) and ② optimizing it (profiling, benchmarking, rewriting). Every line of work has a real cost: time, attention, opportunity cost. You can literally explore combinations of inputs that will never occur, or speed up something nobody will ever use ("you have no users"). Even in a world with an unlimited supply of tokens, you need to know when to stop and which tradeoffs to accept. Agents will happily keep digging under any circumstances.»
Guillermo Rauch
From a tweet@rauchg10/08AI agents need a stopping point: even unnecessary code can be improved forever
«The engineering community will have to rediscover one thing: almost any program can be improved practically forever by ① making it more resilient to failures (fixing edge-case bugs, validating inputs more strictly, handling errors) and ② optimizing it (profiling, benchmarking, rewriting). Every line of work has a real cost: time, attention, opportunity cost. You can literally explore combinations of inputs that will never occur, or speed up something nobody will ever use ("you have no users"). Even in a world with an unlimited supply of tokens, you need to know when to stop and which tradeoffs to accept. Agents will happily keep digging under any circumstances.»
Guillermo Rauch
From a tweet@thdxr10/08
Share a local site with a link: OpenTunnel encrypts traffic all the way to your machine
«opentunnel — public URLs for anything running on your machine tunneling services already exist, but this one is different - end-to-end enc…» — dax
From a tweet@thdxr10/08Share a local site with a link: OpenTunnel encrypts traffic all the way to your machine
«opentunnel — public URLs for anything running on your machine tunneling services already exist, but this one is different - end-to-end enc…» — dax
From an articlearstechnica.com10/08
Adobe alternatives built with AI: Artcraft offers 7 free apps in early alpha
From a tweet@emollick10/07
ChatGPT answers are going interactive: OpenAI rolls out Intelligent UI to everyone
From an announcementAnthropic10/07
Claude Haiku 5.5 for low-cost, high-volume tasks: input costs $0.10 per million tokens
From an announcementAnthropic10/07Claude Haiku 5.5 for low-cost, high-volume tasks: input costs $0.10 per million tokens
▶
From a tweet@claudedevs10/07
Claude gets easier to connect to a browser: the SDK now runs the action loop itself
«The computer use and browser use toolsets are now built into the Claude SDKs for Python and TypeScript. The API tells you where Claude wan…» — ClaudeDevs
From a tweet@claudedevs10/07Claude gets easier to connect to a browser: the SDK now runs the action loop itself
▶«The computer use and browser use toolsets are now built into the Claude SDKs for Python and TypeScript. The API tells you where Claude wan…» — ClaudeDevs
From a tweet@github10/07
Run Copilot commands in isolation: local sandbox available at no extra cost
«GitHub Copilot local sandboxing is now generally available 🙌 Copilot can run commands in an isolated environment with controlled access to files, the network, system capabilities and credentials. You can fine-tune what it is allowed to do. Enterprise teams can also manage policies centrally. Try it in GitHub Copilot CLI, GitHub Copilot app or @code. https://github.blog/changelog/2026-10-07-local-sandboxing-for-gitHub-copilot-now-generally-available?utm_source=x-launch-local-sandboxing-general-availability&utm_medium=social&utm_campaign=oct-7-event-oct-2026»
GitHub
From a tweet@github10/07Run Copilot commands in isolation: local sandbox available at no extra cost
«GitHub Copilot local sandboxing is now generally available 🙌 Copilot can run commands in an isolated environment with controlled access to files, the network, system capabilities and credentials. You can fine-tune what it is allowed to do. Enterprise teams can also manage policies centrally. Try it in GitHub Copilot CLI, GitHub Copilot app or @code. https://github.blog/changelog/2026-10-07-local-sandboxing-for-gitHub-copilot-now-generally-available?utm_source=x-launch-local-sandboxing-general-availability&utm_medium=social&utm_campaign=oct-7-event-oct-2026»
GitHub
▶
From a tweet@runwayml10/07
Request and refine videos in one chat: Runway connects to ChatGPT Astra
«Describe the task. Let it work. Make revisions. All in one chat window: Runway is now right inside ChatGPT Astra. Get started at the link…» — Runway
From a tweet@runwayml10/07Request and refine videos in one chat: Runway connects to ChatGPT Astra
▶«Describe the task. Let it work. Make revisions. All in one chat window: Runway is now right inside ChatGPT Astra. Get started at the link…» — Runway
From a tweet@mitchellh10/07
Transfer files with a call: OpenAI Dot carried out a task across two Macs
«Just called Chungus (my OpenAI Dot) while picking up Mexican food and asked it to transfer last night's Rex benchmark results from my Mac Studio to my MacBook Pro via Tailscale. Got back to my desk—the files were already there. What a fucking time to be alive.»
Mitchell Hashimoto
From a tweet@mitchellh10/07Transfer files with a call: OpenAI Dot carried out a task across two Macs
«Just called Chungus (my OpenAI Dot) while picking up Mexican food and asked it to transfer last night's Rex benchmark results from my Mac Studio to my MacBook Pro via Tailscale. Got back to my desk—the files were already there. What a fucking time to be alive.»
Mitchell Hashimoto
From an articlerswebsols.com10/07
Employees move from Claude to in-house tools: Meta and Microsoft cut usage
From an articlewired.com10/07
AI agent drives a Toyota to a burger joint: only GPT-6 Astra finishes the route in DrivingBench
From an articlewired.com10/07
AI agent drives a Toyota to a burger joint: only GPT-6 Astra finishes the route in DrivingBench
From a tweet@openrouter10/07
Clef on OpenRouter skips text generation: returns decisions with probabilities
«Cloudflare's Clef decision-making models are available on OpenRouter Clef (27B) and Clef Flash (9B) from @Cloudflare are open-source decision-making models. Send text, JSON or images and get typed answers with probabilities instead of generated text $0.24 per million input tokens for Clef, $0.09 for Clef Flash. Output is free http://openrouter.ai/cloudflare»
OpenRouter
From a tweet@openrouter10/07Clef on OpenRouter skips text generation: returns decisions with probabilities
«Cloudflare's Clef decision-making models are available on OpenRouter Clef (27B) and Clef Flash (9B) from @Cloudflare are open-source decision-making models. Send text, JSON or images and get typed answers with probabilities instead of generated text $0.24 per million input tokens for Clef, $0.09 for Clef Flash. Output is free http://openrouter.ai/cloudflare»
OpenRouter
From an articletechcrunch.com10/07
ChatGPT for Teens urges teens to keep talking: even in crisis tests
From an articletechcrunch.com10/07ChatGPT for Teens urges teens to keep talking: even in crisis tests
▶
From a tweet@lovable10/07
Discuss Lovable projects by voice: Chats responds and saves the transcript
«You can now talk to Lovable about anything. Think out loud (literally: we support voice conversations). The chat deeply understands the co…» — Lovable
From a tweet@lovable10/07Discuss Lovable projects by voice: Chats responds and saves the transcript
▶«You can now talk to Lovable about anything. Think out loud (literally: we support voice conversations). The chat deeply understands the co…» — Lovable
▶
From a tweet@higgsfield_ai10/07
Adapt a video ad to your product: Ad Multiplier preserves editing and motion
«Introducing Ad Multiplier in Ads Studio. Turn any ad that already performs well into new data-driven creatives for your brand. > Analyzes…» — Higgsfield AI 🧩
From a tweet@higgsfield_ai10/07Adapt a video ad to your product: Ad Multiplier preserves editing and motion
▶«Introducing Ad Multiplier in Ads Studio. Turn any ad that already performs well into new data-driven creatives for your brand. > Analyzes…» — Higgsfield AI 🧩
From a tweet@dhh10/07
Agents don't need complex setups: DHH gets by with almost no skills
«Exactly. People keep asking me how I have everything set up. It all comes down to any agent harness, several agents running at once, almost no skills and critical review. There's no secret recipe (or secret source code!). The models work great out of the box.»
DHH
From a tweet@dhh10/07Agents don't need complex setups: DHH gets by with almost no skills
«Exactly. People keep asking me how I have everything set up. It all comes down to any agent harness, several agents running at once, almost no skills and critical review. There's no secret recipe (or secret source code!). The models work great out of the box.»
DHH
▶
From a tweet@lennysan10/07
Building products is still hard: Linear CEO sees no breakthrough from AI
«Linear CEO @karrisaarinen: “This summer, I stopped following AI for a while. I thought that when I came back, I'd feel like I'd fallen beh…» — Lenny Rachitsky
From a tweet@lennysan10/07Building products is still hard: Linear CEO sees no breakthrough from AI
▶«Linear CEO @karrisaarinen: “This summer, I stopped following AI for a while. I thought that when I came back, I'd feel like I'd fallen beh…» — Lenny Rachitsky