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
All videos
Full episodes, transcripts, and the stories behind them.
7:22Block Says Dessa Learned From Speech Alone | AI News Briefing
Block says its Dessa voice AI learned from conversation audio alone, while MIT engineers demonstrated a light-steered swimming robot powered by living muscle cells. Grok Imagine’s first Odyssey episode is not the full-length historical film Musk pledged, and a researcher recommends family code words after a scam involving a voice believed, but not confirmed, to be AI-generated.
Watch episode
6:57Shared AI Agents, Meta’s Hackathon and Banks’ New Limits
Sakana AI’s Aquarium gives multiple agents shared business knowledge and controls, while Anthropic says leaner agent setups may be faster but need careful regression checks. Meta and Cerebral Valley plan a virtual AI hackathon, and Axios reports major U.S. banks are limiting agents’ account actions, though Reuters could not verify the specific cases.
Watch episode
8:02Claude Duet Works Between Chats; Math Proof, Jobs Data and PHP Flaws
Anthropic’s Claude Duet can keep work moving between chats, with approval required for consequential actions, while Microsoft researchers describe human-guided AI’s role in proving a decades-old math conjecture. Stanford’s dashboard flags declines in AI-exposed jobs without showing AI caused them, and PHP has patched flaws researchers used to demonstrate server-takeover risks.
Watch episode
8:20ChatGPT Adds PIN-Locked Chats—But Not Encryption | AI News Briefing
ChatGPT is rolling out PIN-protected conversations that disappear from regular history and search, without adding end-to-end encryption or changing data controls; AI solver Darmok beat the human winner in the Times cryptic final. The Boston Symphony will premiere a work using an MIT-built system to connect live performances, while Eric Schmidt reportedly owns an Anthropic stake potentially worth billions—not a confirmed payout.
Watch episode
7:35AI Facial Signals, Corporate Breaches, Scam Decoys & Claude’s Island Skills
A new study finds facial-image AI tools capture complementary signals linked to cancer survival, while Reuters reports an AI-assisted hacker breached sites at more than 100 companies—with the extent of data theft unclear. Apate says its AI decoys have engaged scammers, though their impact needs independent analysis, and a not-yet-peer-reviewed MIT study found Claude agents improved at island tasks after exploration, without proving genuine play.
Watch episode
7:46AI News: Anthropic’s Junior Hiring Shift, ChatGPT Slides, Minecraft Agents & Quantum Circuits
Anthropic says junior hires make up a smaller share of technical recruiting at its highly automated company, while ChatGPT adds collaborative decks with PowerPoint and Google Slides export for eligible users. DeepMind researchers report four agents beat Minecraft in a controlled run, while Anthropic says Claude cut gate counts on quantum-circuit tasks—not runtime.
Watch episode
7:55Claude Opus 5.5 Fast Mode Opens Up—and AI News Across Apple, Google and Microsoft
Anthropic removes the waitlist for Claude Opus 5.5 fast mode, while Apple shares linked-Mac AI results whose gains depend on connection speed and workload. Google demos Gemini 4 Argon for robot control, simulation and training-data review, though reliability remains unclear, and Microsoft pairs junior engineers with senior mentors and university courses for AI-assisted development.
Watch episode