Social Media Briefing — June 15, 2026

Warns that a major leap in Chinese open-weight model performance could trigger a full ban of the Chinese LLM sphere by national security agencies.

Mollick relays a DeepMind researcher finding that when one model trains the next, the successor can inherit hard-to-filter quirks, possibly explaining why models within a family feel similar.

Chollet contends near-term AI is the newest form of digital leverage, a force multiplier requiring humans in the loop at every level to be useful.

Marcus cites a new study claiming that if an AI cannot apply an abstract lesson to a new situation it is not truly reasoning, framing it as backing his 25-year argument.

Delangue frames AI's future as a choice between closed-source APIs concentrating power and open-source AI letting everyone participate, citing an org like the city of Rio.

Argues Anthropic has harmed AI governance discourse but the administration's actions are worse, urging action before stronger models arrive.

Mollick stresses that best practices for rebuilding companies around AI agents are unknown, agents are only months old, and experimentation with productive failures is needed.

Asks how the US will handle a wave of Mythos-level open-weight models in 6-12 months, suggesting a possible full AI ban.

Following yesterday's News of the Fable shutdown, a firsthand look from the ground, Scobleizer gives a firsthand account of Anthropic's Claude Builder Day hackathon in San Francisco, noting it was meant to celebrate a launch but was dampened by Fable being pulled the prior night; builders praised Fable over other models and were disappointed they could no longer use it.

Ethan Mollick highlights a methodological thread debating a paper claiming generalist models outperform specialized medical AIs, and the difficulty of benchmarking AI in medicine.

Suggests recent events mark the start of a tumultuous AI policy era and powerful models could face bans with no defenders.