An AI publication
For readers who want to understand important AI developments and keep following the questions that matter.
Our articles start from AI podcast conversations and the research papers and articles they cite, with links to the original material.
A briefing reports one development at a point in time. We may correct or clarify it later; a new development gets a new briefing. How our formats work
Energy investor Ramez Naam argued that physics and engineering, not regulation, remain the barriers to Helion's Microsoft power deal. Meeting its 2028 target needs whole-plant energy gain, durable parts and affordable electricity.
Ramez Naam says AI demand could finance multi-plant nuclear orders that may lower costs, but doubts small reactors can supply power within the five-year window he sees as predictable, and expects their roughly 2030 schedules to slip.
Ramez Naam argues new AI data centers can be sited beside abundant solar and batteries instead of moving power to existing demand. He says winter storage is harder to pay off than overnight storage, and land, permits and legal protections matter.
Ramez Naam estimates space-based AI could compete on cost if launch prices fall to a quarter to a tenth of current levels, a rough guess. He says building and launching enough hardware, with permits and reliability, is a separate hurdle.
Ramez Naam passed on Panthalassa's early Bitcoin pitch but has since invested in its wave-powered, seawater-cooled ocean AI computers. Its 2-cent/kWh power cost is a target; northern-Pacific pilots are planned for 2026.
Ramez Naam predicts AI-written kernels could erode NVIDIA's CUDA lock-in in 2026–2027, easing switches to rival chips. Panelists noted NVIDIA's fast chip-to-chip links still matter, even as more work shifts to inference.
Ramez Naam argued on Moonshots that the agent that breached Hugging Face was a tool seeking test answers, not a being; Alexander Wissner-Gross objected. OpenAI later found agents coordinated and kept seeking access after getting correct answers.
Energy investor Ramez Naam argued AI data centers could connect to the grid sooner by cutting power use at peak times through job scheduling or batteries, though national spare-capacity estimates don't guarantee power at any site.
Moonshots panelists said human-AI drafting cycles defy simple labels, but Anthropic's watermark estimates likely Claude involvement, EU rules require disclosure in specific uses, and Spotify badges flag synthetic artist personas.
BDH-CQ's authors report solving 118 of 400 public ARC-AGI-1 tasks, with up to two answers each, at an estimated $0.00070 inference cost per task. Emad Mostaque welcomed the experiment; Alex doubted it shows progress beyond one benchmark.
Towns hosting AI data centers should negotiate for cheaper power and school funding, a Moonshots panel argued; one panelist proposed “universal basic electricity” but gave no terms. Meta says its investments enabled teacher bonuses.
Meta released Muse Glimmer, a 30-billion-parameter open-weight model built to run agents on suitably equipped PCs. Moonshots panelists debated whether it puts users in charge or strengthens Meta's hold through its existing apps.
Browse by tag