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
Box CEO Aaron Levie argues AI spread fastest through coding because code is text and engineers fix their own tools, while jobs like sales lag. He is betting, admittedly with bias, that the money lies in bringing AI into other workflows.
Professor Jiang says his main worry about AI is not surveillance but that it cheapens the human experience and distracts people from spiritual growth, though he allows AI may outdo humans at mathematics and medical diagnosis.
Professor Jiang argues technology rests on "specialization times globalization"; if war breaks global supply chains, he says, AI will not reach AGI but be repurposed as a surveillance control grid. The host expects that surveillance anyway.
Alexander Wissner-Gross argued next-word prediction as a training task may last indefinitely, while the transformer architecture is already being replaced piece by piece. Another panelist guessed chips move to photonics in 18 months to two years.
Asked by an educator what to teach when future jobs are unknown, Salim and Peter urged starting children on a problem and purpose rather than a profession; another panelist warned kids still need school to become healthy adults.
The Moonshots AMA panel told caregiver-founder John Regan that exhausted family caregivers should start with AI day planning, then voice systems, and only later robots. No one on the call reported results from the approach.
Moonshots hosts told mid-career professionals with mortgages to brainstorm spare-time AI experiments with a language model, tinker at night, find online peers, try Mercor and schedule slack time; none of it shows an income can be replaced.
Data buyers aren't haggling but overwhelmed, a Moonshots panel told a civil engineer whose 30-year archive drew interest from Mercor and Scale but no deal after weeks. They advised packaging it with his own price.
On the Dwarkesh Podcast, AI researchers argued that pass/fail rewards work because earlier training supplies candidate behaviors and rewards ignore incidental wording. Charlie O'Neill warned that harder training tasks cost ever more to build.
John Schulman expects frontier AI models to keep growing as compute and GPUs expand. Charlie O’Neill expects little parameter growth for a few years, because long reinforcement-learning runs reward cheap inference.
Charlie O'Neill told the Dwarkesh Podcast that AI already learns from deployment, but in slow batches, and that many small updates erase earlier learning; John Schulman expects customers to resist feeding a shared model.
AI researcher John Schulman says cheap distillation keeps model competition open. He argues that a copy gets broad behavior only from realistic, multi-turn prompts, and that benchmark-style tasks alone may match scores without it.
Browse by tag