Weights & Words Reading Group x AI Debate

​Join us for a reading group and live debate on scaling laws — and what's left of them.

​Distilled community is teaming up with AI Debate group for this one!

​A lot has happened this year. Frontier compute is still climbing roughly 3.5× per year, but the real constraints have moved to power, debt, and data. Sparse MoE made the data wall worse, not better. RL got its own "scaling laws," which may not deserve the name. METR's time-horizon suite hit its measurement ceiling in March. Labs started writing "helped build itself" into release notes. And for the first time, a frontier model was withheld from release because of what it could do.

​The format is discussion and debate, not lectures. We'll open with a short framing of where things stand, then move into structured debate. Each motion is a single sharp claim — "withholding frontier models for “safety” is becoming a marketing strategy," "the first lab to automate its own research won't tell anyone," "AI companions make better partners than people" — and you'll be asked to argue a side, not necessarily your own. Some motions are technical, some are about where all of this leaves the rest of us. A crowdsourced prompt box will be open for attendees to submit their own.

​This is for people who train, evaluate, study, or just think hard about these systems. Expect disagreement, expect to be wrong at least once, and bring a plot or a take you think everyone is misreading.

​Note on RSVPs. Capacity is intentionally limited to keep the room small and conversational. If your plans change, please update your RSVP so we can open your spot to someone else.