Quiet slot. The signal is one tweet relaying Helen Toner's sworn testimony that OpenAI's Deployment Safety Board was structurally tilted toward Microsoft plus Altman.
Curated reposts
No retweets in this slot.
AI Account Feed
- OpenAI Deployment Safety Board structure exposed under oath (@ns123abc). Helen Toner testified that the DSB had three Microsoft seats and three OpenAI seats with majority approval needed, and that Sam Altman held one of the three OpenAI seats. Three Microsoft votes plus Altman equals four, the threshold for deployment, even if both other OpenAI safety members voted against. The formal pre-deployment safety review was a rubber stamp for commercial release. Pairs with the Musk trial coverage in 05-06.
- Speculative xAI / Anthropic Colossus 1 lease economics (@ns123abc). Reads as a thought experiment, not a report. Premise: xAI's Colossus 1 is at 11% utilisation after frontier training moved to Colossus 2, Anthropic is capacity-constrained, the rumored 220k-GPU lease at 87% cash margin gets pledged as collateral for new debt. The numbers are illustrative; the pattern (training fleets depreciate fast once they go off-policy, inference demand amortises them) is the real signal. Sits next to the 05-06 reporting on the actual Anthropic-Colossus deal.
- Lex Fridman FFmpeg podcast (cluster of 2) (@lexfridman). Two near-duplicate promo posts for an FFmpeg / VLC episode. Skip.
Connections
- The OpenAI DSB disclosure is the procedural counterpart to the Musk-trial distillation admission from 05-06. Both tighten the public record on how OpenAI made consequential decisions during the Microsoft-partnership era. The trial is becoming the wiki's primary source for OpenAI governance ground-truth.
- The Colossus 1 lease commentary lines up with the Anthropic/SpaceX Colossus deal from the 05-06 digest. The thought experiment names the unit-economics pattern (lease-and-collateral over pure cap-ex) that frontier-lab infrastructure now runs on, regardless of whether the specific xAI-Anthropic numbers are real.
- @ns123abc is the slot's only AI-relevant signal. Both posts intersect ongoing wiki threads (governance, infrastructure economics) without duplicating headlines already covered. The handle has become a reliable signal carrier for sworn-testimony fragments worth filing.