Sergey Brin on the AI revolution, robotics, education, and the future of human-computer interaction

Threatening language models improves performance.

Across models, adding coercive or threatening prompts yields better outputs.

Why: Likely exploits alignment training that rewards compliance under perceived pressure.

"not just our models, but all models tend to do better if you threaten them... Like with physical violence... historically you just say, oh, I'm going to kidnap you if you don't blah, blah, blah. They actually"

AI capability is advancing faster than any prior technology wave.

Unlike the web or smartphones, AI systems change so rapidly that a one-month absence yields noticeable leaps in performance.

Why: The compounding nature of model improvements and compute scaling creates exponential gains.

"these AI systems actually changed quite a lot. If you went away somewhere for a month and you came back, you'd be like, whoa, what happened?"

Current AI progress dwarfs the early web era.

While the early web grew quickly in adoption, its technical underpinnings evolved slowly compared to AI.

Why: AI models undergo fundamental capability jumps month-to-month, whereas web tech matured over years.

"the developments in AI are just astonishing, I would say by comparison, just because of the web spread, but didn't technically change so much from month to month, year to year"

Pre-training is only the beginning; post-training unlocks reasoning.

Early focus on massive pre-training has shifted to post-training techniques that yield thinking models.

Why: Post-training refines raw model intelligence into practical reasoning capabilities.

"I was really very close with what we call pre-training... More recently, the post-training, especially as the thinking models have come around. That's been another huge step up"

AI’s superpower is volume of work humans cannot match.

Deep research can ingest thousands of sources and perform follow-on queries equivalent to weeks of human labor.

Why: Scale transforms AI from helpful to indispensable.

"I think of the super power is when it can do things in a volume that I cannot... if it sucks down the top 1000 results, and then does follow-on searches for each of those, and reads them deeply, that's a week of work for me"

Traditional schooling is already obsolete for core STEM topics.

AI already surpasses high-school and college-level math and coding; curriculum cannot keep pace.

Why: AI improves yearly while human education is static or slow-moving.

"I see I've a kid in high school... the AIs are basically, you know, already ahead... if you talk about like math or calculus or whatever, like, they're pretty damn good"

College ROI is collapsing under AI pressure.

Even before AI, vocational and cost-benefit questions were rising; AI accelerates the reckoning.

Why: Skills taught may be automated before graduation.

"it seems like college was already undergoing this kind of revolution even before this sort of AI challenge... AI obviously puts that at the forefront"

Humanoid form factor is overrated.

World designed for humans can be learned by AI without mimicking human shape.

Why: AI generalization from simulation and video is strong enough to adapt to varied hardware.

"I personally don't think that's given the AI quite enough credit... I don't know that you need exactly the same number of arms and legs and wheels... as humans, to make it all work"