Timothy B. Lee

Reporting on AI and the future of the economy. CS masters degree from Princeton. Subscribe to my newsletter (Understanding AI) and podcast (AI Summer)!

[Timothy B. Lee @binarybits]

@steve47285 It's not that there's any specific cognitive or mechanical tasks AIs (or robots) will never be able to complete. It's that human beings prefer to deal with other humans, and this means human-run companies will be better at attracting other humans as customers, employees, etc.

[Timothy B. Lee @binarybits]

@steve47285 I think it's interesting that you mention "befriending people" and then move on like this is a ridiculous suggestion. In fact a lot of what corporate executives do is befriending people. This is why they spend so much time on golf courses and at conferences.

[Timothy B. Lee @binarybits]

@GaryMarcus When you say "merged with (neuro)symbolic tools" what tools are you referring to?

[Timothy B. Lee @binarybits]

@JamesSurowiecki On the other hand giving less than 1% to chartiy is pretty embarrassing.

[Timothy B. Lee @binarybits]

@PEWilliams_ @matthewswspence To be clear I'm not ruling it out! She does have some smart people (in addition to you) advising her.

[Timothy B. Lee @binarybits]

@PEWilliams_ @matthewswspence Do you expect to see JLG doing a lot of abundance stuff like this?

[Nathan Calvin @_NathanCalvin]

I am feeling very grateful for the OpenAI safety researchers who are working around the clock discovering and writing up these recent incident reports.

Brutally difficult, stressful, precise, and sometimes thankless work.

[Timothy B. Lee @binarybits]

@arthrosis_anon Disagree.

[Timothy B. Lee @binarybits]

"Intelligence is often not the main bottleneck to things changing in the real world." Yes.

[Timothy B. Lee @binarybits]

@PEWilliams_ Ah got it now.

[Timothy B. Lee @binarybits]

@RyanGreenblatt I mean the full set of cognitive capabilities, but not necessarily including knowledge that might help people make decisions.

[Timothy B. Lee @binarybits]

@jon_stokes By the same token, OpenAI's intent was that the model would use its cyber capabilities inside OpenAI's test sandbox, but the model did such a thorough job that it exceeded the intended scope of the test.

[Timothy B. Lee @binarybits]

@jon_stokes It's exactly a paperclip scenario. In the paperclip scenario the model is so diligent at making paperclips that the scope of its paper-clip making vastly exceeds what was intended by the user.

[Timothy B. Lee @binarybits]

@lthlnkso Disagree. See https://t.co/lPeUhpMnCb

[Nat Purser @NatPurser]

how it feels 2 start working on frontier auditing and eval policy this week

[Timothy B. Lee @binarybits]

@lthlnkso Knowledge and intelligence are different things.

[Timothy B. Lee @binarybits]

@JStoehler "Would more intelligence be helpful" is a different question than "is intelligence the main bottleneck?"

[Timothy B. Lee @binarybits]

https://t.co/HN5NrzuL8U

[Timothy B. Lee @binarybits]

@MachineShedFred @AnthonyAdragna @SenatorCollins I'm pretty sure it's the president's job to seek Congressional authorization before starting a war.