H-1B Shakeup, Kimmel Apology, Autism Causes, California Hate Speech Law

Trump’s new $100,000 one-time H-1B fee was framed as a market filter meant to reserve scarce visas for genuinely hard-to-replace talent. Sacks and Chamath argued that outsourcing-heavy use and lower pay have distorted a program capped at 85,000 new visas, while OPT and national-interest waivers could preserve access for elite graduates and researchers.

Friedberg argued the U.S. should separate mass immigration from strategic talent recruitment and actively target a few thousand world-class AI, chip and scientific specialists. The proposed model resembles a modern Operation Paperclip, using government and industry to recruit scarce people whose location could materially shift technological competition.

The strongest autism discussion centered not on a single cause but on a measurable folate-receptor autoimmune pathway that may explain a subset of cases. Friedberg said an auto-antibody test has existed since about 2012, cited one study finding it in over 70% of a severe sample, and noted planned leucovorin labeling work.

The Tylenol evidence presented was association, not proof of causation. A 46-study review reportedly found 27 positive associations, nine none and four negative; its Harvard author had also been a $150,000 paid expert against Tylenol and later stressed that the link remains uncertain.

An MIT/Microsoft planning paper reportedly pushed Llama 3 from 1% to 64% on a planning benchmark by training explicit state-action-state reasoning with external validation. The framework reached up to 94% on some benchmarks, suggesting large gains from teaching models structured planning rather than relying only on raw next-token prediction.

A German paper claimed an attention-memory architecture could deliver 100× faster inference and 70,000× lower energy use than an H100 in its tests. Friedberg said edge AI could become dramatically more practical if it scales; Sacks cautioned that academic breakthroughs matter only once reproduced in products.

The All-In “shadow ban” turned out to be a profanity-detection bug: muted swear words remained in YouTube’s transcript and triggered Restricted Mode. That silently cut traffic on networks using Safe Mode; YouTube told the hosts mass reporting was unrelated and said creators currently lack a clear restriction notice or reason code.

Friedberg said YouTube has now acknowledged being pressured by the Biden administration to remove content, placing it alongside similar claims previously made about Meta and Twitter. The hosts connected that history to California SB 771, arguing that vague hate-speech liability could again pressure platforms to suppress lawful but offensive speech.