The AI Preference Cascade Reaches Farther

Robert O'Callahan announced he resigned from Google, writing that while he enjoyed his work and loved the people, his GDM team was working on a new generation of chips intended to make AI much faster and cheaper, and that he believes AI is already progressing too fast.

According to the account, Palisade Research released a batch of interviews with current and former lab employees about their personal views on existential risk. The account states these interviews tend to come from former employees rather than current ones, saying current employees are discouraged from discussing the topic, primarily through cultural pressure and sometimes a lack of permissions, especially at Anthropic.

The account argues this counters claims that such warnings amount to marketing or regulatory capture, reasoning that AI companies pushing such narratives would not censor their own employees on the same topics, and that the most concerned employees would not feel the need to resign in order to speak.

The account also describes a shift in public opinion, saying people are following the warnings, though it expresses skepticism that the full extent survives adjustment for salience. It states that slowing down AI development is very popular relative to the pace voters expect, and that salience, while rising fast, remains low for now.

The account frames these events as a continuing preference cascade, saying it kept accelerating after the HuggingFace incident and reached a tipping point with the Coxon resignation, and that it shows no signs of stopping.