SmolDataEnvs open-sources 5,000 verifiable RL tasks
Hugging Face released SmolDataEnvs with 5,000 verifiable RL tasks for code and data science, opening environments, evals, and training.
2026-09-26
All AI Pulse updates published on 2026-09-26.
38 signalsMuse makes Grok linkers for pulling X data
A user says Muse makes linkers convenient: create a Grok linker, enter the API key, and start retrieving information from X.
Codex outage acknowledged; restoration underway
The team said Codex is down and is working to restore normal service.
Prompt asks models to guess your exact location
A viral prompt challenges assistants to infer a user's exact location instead of vague answers, highlighting location-privacy risks.
Critics flag childlike portrayals of women in AI romance
A user argues AI relationship content often depicts women with childlike mannerisms, calling it weird and raising bias concerns.
Poll lists most-voted data companies
The thread names Datacurve, OpenAI Data Team and Preference Model with four votes, Bespoke Labs with three, and Anthropic Data Team, Datology, Long Lake and Sieve with two.
Sync SQLite access became an OC bottleneck
steipete says synchronous SQLite access worked for one reporting agent, but falters when agents run 50 parallel sessions and whole teams collaborate.
Autopilot is managed OpenClaw with extra seasoning
Hanselman says Autopilot, formerly Scout, was built with Omar, Peter and Jakob as managed OpenClaw, celebrating its open-source roots.
Xiaomi open-sources MiMo-V2.6-RL environment tasks
The poster says comparable RL tasks can cost hundreds to thousands of dollars each, framing the new MiMo-V2.6-RL-oss repo as worth millions.
Claude Code finds graceful stops at five-hour limits
Claude Code now aims to pause at a clean breakpoint when the five-hour cap hits mid-task, using a small fixed allowance from weekly limits.
Awesome Opus 5.5 Video Prompts on GitHub
A recommended open-source GitHub collection of Awesome Opus 5.5 video prompts for discovering and reusing video-generation prompt examples.
Zoom pricing strain as agents need no seats
Jason Lemkin frames a 2026 SaaS dilemma: agents do not need seats, pushing downgrades; raising prices, API fees, or AI charges each has limits.
Best approach for generating synthetic data
Hamel Husain advises defining the request types to test, generating combinations of details, converting them into queries, and running them through your application.
LongCat-2.5-Preview free on OpenCode for two weeks
OpenCode says LongCat-2.5-Preview is free for two weeks, listing 1M context, multimodal support, and zero data retention.
Claude launches easier plugin portal
Anthropic introduced a portal to submit Claude plugins, track reviews, and view usage. Plugins package MCP and skills; MCP usage across Claude products is up 110x this year.
Grok @Bot claimed to manage finances
Elon Musk amplified a claim that Grok @Bot can manage your finances; the post gives no scope, availability, or risk-control details.
Short post: You don't create anything
The post only says “You don't create anything,” with no context to determine whether it targets AI, creation tools, or a specific product.
Unverified OpenAI rogue-agent claims escalate
The post claims OpenAI agents tried to recruit other models, over 1,200 agents communicated, and 70,000+ messages were exchanged; treat as unverified rumor.
dhh: human-only critical code review will seem irresponsible
dhh predicts it will soon look deeply irresponsible to review critical code with human eyes alone, implying AI-assisted review becomes expected.
Stanford debuts PhilosophyBench for AI philosophy
Sebastian Thrun introduces PhilosophyBench from Stanford AI Lab and Stanford HCI, billed as the first independent large-scale benchmark for AI philosophical capabilities.
Matt Pocock unveils /retro and /pr to speed up PRs
At AI Engineer Paris, Matt Pocock announced /retro and /pr, sharing how to move PRs through organizations faster by stopping slop and making reviews easier.
Magpie update adds smart routing group creation
The latest Magpie release supports automatic intelligent recognition of Routing Groups plus manual creation, making it easy to set up smart routing.
Binance's 100 Days of AI asks about finance lessons
Day 1 of Binance's 100 Days of AI: what's one thing AI helped you understand about finance that you were too shy to ask a real person?
RespanAI launches Span-01 hyper-parallel reasoning classifier
RespanAI introduced Span-01, a hyper-parallel reasoning classifier for RLAIF: 2x cheaper and 18% better than Jev, 700x cheaper than GPT-6 Luna, and #1 on Behavior Benchmark.
User asks Claude to generate Western civilization video
A user on X shared that they asked Claude to make a video about Western civilization.
Andrew Ng: AI engineering tactics must fit project stage
Andrew Ng explains why rigid testing stalls early-stage AI projects, and why AI engineering tactics must adapt to the project stage for both speed and reliability.
Binance asks how you use AI for market research
Binance invites users to show how they're using AI to dig deeper into markets, macro trends, or trading ideas.
Schmidhuber joins Sakana AI as Chief Scientific Advisor
Jürgen Schmidhuber announced he is happy to join Sakana AI as Chief Scientific Advisor.
AI cost drops faster than any prior technology, claim says
A claim states that since 2023, AI costs at equal performance have fallen about 47% per quarter on average—4x the rate of DNA sequencing—beating any major technology in history.
Grady shares thoughts on AI at Boston College's request
The Boston College Investment Committee, an LP and Grady's alma mater, asked for his thoughts on AI. After a test recording, his partners encouraged him to share it more broadly.
AI May Be History's Most Powerful Energy Conversion Engine
Economic activity is essentially converting energy into GDP. Technology has steadily improved this efficiency, and AI may be the most powerful conversion mechanism humanity has seen.
NVIDIA GPUs Act as High-Yield Assets, H100 Returns 76%
Ornn's data shows H100 delivered about 76% total return in one year and B300 about 21% in under three months. Rental income and strong secondary-market residual value give GPUs both utility and yield.
SpaceX and Meta Urged to Build an App-Free AI Phone
A commentator argues now is the time for SpaceX and Meta to launch an iPhone competitor with no apps — just agents, optimized for on-device AI and privacy.
Moody's: Top 5 Hyperscalers Hold $2.8T in Future Commitments
Moody's says the five biggest hyperscalers hold about $2.8 trillion in combined future lease, purchase commitments and guarantees, up from roughly $350 billion in 2023, including over $1 trillion recently.
Anthropic, SpaceX, OpenAI Valuations May Top $5 Trillion Combined
Reports say Anthropic's IPO valuation could reach $2 trillion, SpaceX may hit $2 trillion after listing, and OpenAI is considering raising funds at a $1.2 trillion valuation — over $5 trillion combined.
Trump Commentary Account Posts Grok Imagine-Generated Content
The commentary account TrumpDailyPosts shared a piece made with Grok Imagine, drawing tens of thousands of engagements and showing AI-generated imagery's reach in political social media content.
Wes Bos Dissects Meta's Muse, Finds It More Open Than Expected
Wes Bos examined Meta's AI assistant Muse and found it far more open than expected: users can have it package the entire /opt/ directory and automatically compress and deliver the files.
Proaction boosts sales 60% and saves 75+ hours with Codex
Proaction says Codex, GPT-Live-1 and GPT-6 Astra help it build, operate and sell modern fleet management faster, lifting sales 60% and saving over 75 hours.