Anthropic co-founder Jack Clark says AI systems will contribute to a Nobel Prize-winning discovery within 12 months and design their own successors by the end of 2028

Anthropics Jack Clare is super optimistic:

- End of 2028 "AI system would be able to design their own successors"

- within 12 months humans + robots will find nobel prize-winning discoverys

From his lecture at Oxford two days ago.

https://www.theguardian.com/technology/2026/may/21/ai-nobel-prize-winning-discovery-robots-jack-clark-anthropic

Thanks to @curiousiter, here's a slide from the lecture:

@deredleritt3r conservative imo

Claude loves to use 'vertiginous' by the way.

@deredleritt3r If AGI truly comes to being, I hope it will be more like this scene from Peaky Blinders, not the doom and gloom type.

A divine arbiter, not just another being fighting for the resources.

@deredleritt3r Nobel - sure, could happen. Robot tradesmen - more than 2 years, real world moves slow RSI - 2.5 years feels long Fully run million dollar AI company - Maybe? I feel like even a tiny error rate could prevent this.

@deredleritt3r This appears to be a photo of the predictions from someone at his lecture:

@deredleritt3r @MudgilShreshth 'Jack Clark described a “vertiginous sense of progress” in the technology and made a series of predictions'

Claude loves to use 'vertiginous'.

@DualAcies @deredleritt3r I always hope it's gonna be something like this to us

@deredleritt3r Sure, sure, but when will AI be able to reliably count and label things?? Or are we just doing Nobel prizes now, counting later?

@AndrewCurran_ Agreed with nobel prize-winning discovery in 12 months.

2027 will be interesting. It will be important to be able to direct swarm intelligence and getting them to stick to plan.

Initially, bots will be handled for repetitive and dangerous parts of work then later take on rest.

@AndrewCurran_ Claude absolutely loves the word vertiginous lol

@curiousiter Thank you!

@lightconefiller @deredleritt3r I think the problem is that humans tend to correct more effectively, and they tend to not make errors as obviously wrong as AIs do.

I could see an AI committing an error that lead to a bankrupting lawsuit or being defrauded of all its resources in ways that humans would not.

@kimmonismus The interesting part is not just whether the prediction is too optimistic

It’s that AI is moving from answering questions to helping humans explore, test and discover faster

That shift is much bigger than chat

@losslandscape @deredleritt3r re: error rates, how do you compare to human companies? Because it seems to me that existing human companies have non-trivial error rates yet they persist due to error correction and general resilience (most errors are small and easily fixed)

@deredleritt3r Interesting set of projections. Even if timelines are uncertain, the direction of tighter human AI coupling in research and industry feels clear.