🕶️ AlphaGo founder: AI learns by doing

Tuesday’s operating theme is simple: AI is reorganizing around control. David Silver just raised $1.1B to build ‘superlearners’ that learn from simulation, not the internet; meanwhile, OpenAI, Microsoft, Musk, and Altman are busy turning governance into a product surface. Add Beijing killing Meta’s $2.5B Manus deal on national security grounds, and the mood shifts from acceleration at all costs to acceleration with legal, geopolitical, and cap-table paperwork.

Even the robots are being trained for uncertainty. Quite a roadmap.

Power users figured this out early: speaking a prompt gives you 10x more context in half the time. You include the edge cases, the examples, the tone you want — because talking is fast enough that you don't skip them.

Wispr Flow captures everything you say and turns it into clean, structured text for any AI tool. Speak messy. Get polished input. Paste into ChatGPT, Claude, Cursor, or wherever you work.

89% of messages sent with zero edits. 4x faster than typing. Works system-wide on Mac, Windows, and iPhone.

Start flowing free

Atech, a Copenhagen AI hardware startup, raised an undisclosed pre-seed to build a “vibe‑engineering” platform that turns plain‑English prompts into working device prototypes. The round includes Nordic Makers, Emblem, Lovable, and the Sequoia and a16z scout programs, and comes with a direct endorsement from Lovable’s CEO. Tech press detailed that it raises pre-seed funding and receives scout backing, while Atech says its platform will build natural-language prototypes. The pitch frames Atech as a hardware analog to prompt-built software tools and slots into the rising “Physical AI” wave.

The least obvious ripple is upstream: certification labs, component marketplaces, and contract manufacturers could see earlier, smaller orders from non‑specialist builders as prototyping speeds up. If Physical AI keeps accelerating, the edge shifts from who can design to who can source, certify, and assemble at short notice.

Founders in robotics, IoT, industrial automation, and consumer devices should plan for prototype cycles collapsing from months to days, then pressure‑test where abstraction ends, PCB layout, component availability, firmware reliability, and design‑for‑manufacture. Also note that scout participation is signal, not a guarantee of follow‑on, so secure manufacturing and compliance partners early while you validate that a prompt‑to‑prototype actually survives real‑world testing.

David Silver is back with a new bet: that the path to superintelligence isn’t more text, it’s more experience. His startup just raises $1.1B seed at a $5.1B valuation to build “superlearners”; reinforcement learning agents that teach themselves inside rich simulations instead of training on human exhaust.

If LLMs are “fossil fuel,” as Silver puts it, Ineffable’s thesis is that the next frontier is “renewable” learning systems that can compound capability without new labeled data. That flips the usual AI playbook: the core asset isn’t proprietary corpus access, it’s the quality of environments, reward design, and access to compute, which Silver has already locks in massive compute.

Superintelligence-by-simulation also makes safety an experimental variable, not a prompt-engineering afterthought. Silver’s lab is effectively a controlled test of whether one can bets on self-learning RL and still keep the resulting agents aligned.