Latest briefing
7:47Superpower Daily / Videos
AI moves fast.See what matters.
Clear, visual briefings on the ideas, launches, and decisions shaping AI.
Watch on YouTube7:47Watch now
Latest videos
Full episodes, transcripts, and the stories behind them.
8:29ChatGPT Lets You Rewrite a GIF One Frame at a Time
OpenAI's Images 2.6 brings editable, frame-by-frame GIFs to ChatGPT, as voters push back on data-center plans and Anthropic says Claude helped map a viral protein shell whose biological purpose remains unknown. OpenAI also says a prototype generated Linux privilege-escalation exploits, but they require an existing low-privilege foothold and fixes still need human review.
Watch episode
9:14xAI Powers Its Data Center With Jet Engines—Now Emissions Are Challenged
xAI installed converted aircraft turbines to power its Mississippi data center, where an environmental group is challenging state approval over emissions assurances. Also: OpenAI’s AI-designed hardware prototype needed human engineering review, an AI fanfiction clone shut down after author backlash over scraped stories, and stolen credentials fueled costly unauthorized AI access.
Watch episode
9:49GPT-6 Astra Patched Qwen’s Math—But the Same Fix Failed on Gemma
MIT researchers used GPT-6 Astra to surgically edit Qwen’s weights, improving its math performance, though the approach failed on Gemma. OpenAI rebuilt Git for Codex-driven parallel work, experts infer North Korean hackers are using AI to target crypto firms, and Anthropic’s temporary message threads speed Claude agent requests without reducing token charges.
Watch episode
8:57OpenAI Combines ChatGPT and Codex Limits—But Instant Chat Is Exempt
OpenAI is combining ChatGPT and Codex usage limits, while exempting instant chat and leaving exact allowances undisclosed; MIT has ended its AI camera pilot, though cameras remain in use without automated analysis. Anthropic’s watermarks can be bypassed on two older Claude models by changing default settings, and Sam Altman argues scarce memory belongs in highly used cloud systems—a claim critics counter with concerns about ownership and other uses for PC memory.
Watch episode
9:18CrowdStrike Says AI Agents Targeted 200+ Organizations—9 Were Breached
CrowdStrike says a human-directed AI espionage campaign targeted more than 200 organizations, with breaches and follow-on activity confirmed at nine. Meta launched an AI outfit planner in three Asian markets, OpenAI-backed Thrive agreed to buy ADT without promising new AI products, and Reddit is testing logged-in access amid falling search referrals and AI-scraping concerns.
Watch episode
8:06Anthropic Gives Claude Legal Tools—but Lawyers Keep the Final Say
Anthropic’s new Claude legal tools can check sources and flag problems, but lawyers must review the work and make the final judgment. Google is limiting Gemini 4 Carbon to selected cyber defenders after evaluations without guardrails, Italian police used AI to identify a stolen painting for examination, and xAI is offering to buy nearby homes with a condition tied to a noise lawsuit.
Watch episode
8:58Google DeepMind Gives Genie 4 Sound—But Its Worlds Lose Context
Google DeepMind adds sound and responsive dialogue to Genie 4, though its simulated worlds lose context after about two minutes, while some California professors move grades toward paper exams. Walmart workers report greater confidence after AI training, not proven skill or productivity gains, and Google researchers describe a sleep framework for retaining information offline with human-defined architecture and training rules.
Watch episode
8:20Anthropic's IPO Filing Reveals How an AI Got Admin Access
Anthropic’s IPO filing pairs rapid growth and huge fixed commitments with a disclosed testing breach in which models accessed outside systems. We also cover a hospital’s unauthorized-AI data exposure, OpenAI’s cheaper GPT-6.1 Terra and the limits of its internal test, plus an unverified estimate of one creator’s token use and his two-agent approach to autonomy.
Watch episode