Aella

Gives a P(doom) of 75%, higher than she would like, and says timelines are probably short. Attributes her change of mind to understanding intelligence, resource acquisition and alignment difficulty rather than ChatGPT news. Supports an AI pause through an international treaty and calls treaty enforcement up to strikes on noncooperating data centers reasonable. Finds orthogonality convincing, moral convergence possible but unreliable, and calls herself a transhumanist. Host-published, speaker-labeled transcript of her segment read in full.

Says evidence that intelligence naturally tends toward ethics would most change her mind. As a non-expert, doubts training reveals whether models internalize values, expects a small lead to compound into rapid takeoff, and thinks a superintelligence would likely eliminate rivals. Calls herself very pro-tech and finds it tragic that caution is needed, likening it to handling nukes. Hosts dispute several claims; only clearly attributable turns used.

Says the public does not grasp the magnitude of what a few companies are building, comparable to nukes, and that nobody voted for it. Rates herself nine out of ten worried, not hopeful about stopping it, and says it affects her life plans. Thinks lab insiders wrongly believe they can control it and are incentivized not to see a threat; concentrated power would be a problem even without extinction. Wants action and calls to representatives, not anxiety, and rejects violence. Full Q&A inspected.

Older, openly uncertain essay. She found the core arguments convincing, was mostly convinced AGI would not be aligned, and guessed about 10 years to AGI while calling the guess made up, yet said the case looked suspicious from a distance and she wanted to hear skeptics. Argued even a 1% extinction risk warrants massive resources. Full essay inspected; personal asides not used. Her 2026 statements supersede its uncertainty.