Tim H 🦥

econ ph.d. (applied micro) thinking about philosophy of science, AI, mimetic ecclesiology

[Tim H 🦥 @creabuntur]

"'By their very sectarian insistence on the unique truth of Catholicism, these people cut themselves off from access to the truth which they think is theirs, but which is only true when it is received as given.'
https://t.co/d9mpc8CvNg

[Tim H 🦥 @creabuntur]

@krishnanrohit @binarybits @StefanFSchubert @tyler_m_john @bshlgrs @DKokotajlo Michael Strevens has shown the "iron rule" (to base public scientific argument exclusively on empirical tests) works like blinders to impel careful measurement for long-term scientific progress but ought not constrain philosophy or timely policy debate.

[Tim H 🦥 @creabuntur]

@krishnanrohit @binarybits @StefanFSchubert @tyler_m_john @bshlgrs I mean, "falsifiability" is not a helpful criteria. Like "Dunning-Kruger," it only applies to others' views. People like @DKokotajlo have a long track record of predictions you can check. Otherwise, there's no way around reasoning through things, gaming them out.

[Tim H 🦥 @creabuntur]

@binarybits @StefanFSchubert @tyler_m_john @krishnanrohit @bshlgrs I guess Dean is in that category also. But I'd like to see you hash out your differences on these core questions a little more (if you didn't, in the episodes I haven't listened to yet). I find some of your points (in the review, say) persuasive but wonder how they'd respond.

[Tim H 🦥 @creabuntur]

@binarybits Fair enough. Nevermind until I can formulate a better question.

[Tim H 🦥 @creabuntur]

@binarybits @StefanFSchubert @tyler_m_john @krishnanrohit @bshlgrs I'll have to listen to your interview with Ajeya Cotra, but I'm not seeing much else (beyond the review I'd already read). I would accept an invite to The Curve. :o)

[Tim H 🦥 @creabuntur]

@binarybits @StefanFSchubert @tyler_m_john And that's fine. But I wish you and @krishnanrohit would engage more deeply with other worldviews more often. It's fine to reiterate your own and preach to your choirs most of the time, but we need some more sustained back-and-forth (from both sides) to make epistemic progress.

[Tim H 🦥 @creabuntur]

@binarybits @StefanFSchubert @tyler_m_john This is the sense in which I perceive you to have a Hayek-inflected libertarian worldview. You take it for granted that government regulation of AI is a bad idea.

[Tim H 🦥 @creabuntur]

@binarybits Setting aside Chollet's thread, my question to you is whether the capabilities Greenblatt outlines would make an AI swarm more capable than a human at CEO-like tasks of overseeing and coordinating large and complex projects, incorporating many diverse information sources.

[Tim H 🦥 @creabuntur]

@binarybits You implied that scaling up LLMs using methods similar to those behind o1 (and, specifically, without test-time training) wouldn't be enough to ace ARC-AGI and higher-level math. TBF, you don't say that explicitly (and often avoid making predictions the way lesswrong folks do).

[Tim H 🦥 @creabuntur]

@binarybits @Cucho @LeahLibresco @AlanMCole What if they're on a spectrum speeding up exponentially?

[Tim H 🦥 @creabuntur]

@binarybits With Warp Speed being the biggest success story, I guess?

[Tim H 🦥 @creabuntur]

@binarybits Ave what successful examples do you have in mind of rapid US policymaking, responding to a month-by-month development with just-in-time policy?

[Tim H 🦥 @creabuntur]

@binarybits Do you disagree that we're delinquent on capacity building, building expertise within the government and laying the groundwork for compute verification and other monitoring?

[Tim H 🦥 @creabuntur]

@binarybits @bshlgrs @RyanGreenblatt People don't want 10x t shirts, but they want big ass trucks and houses and things. (This is partly a question about whether you expect redistribution.)

[Tim H 🦥 @creabuntur]

@binarybits @bshlgrs @RyanGreenblatt You don't need military spending to dominate GDP to get 15% growth. We literally had 15% growth for two years around 1942.

[Tim H 🦥 @creabuntur]

@rezendi Same question to you then. https://t.co/acwXRgtqIi

[Tim H 🦥 @creabuntur]

@krishnanrohit Would you be willing to go on Doom Debates with @liron and talk through your latest guess at each of the components of your equation?
(Your components have some resemblance to the elements of his "Doom Train" I learned about here: https://t.co/vLhjCYfSz4)

[Delip Rao e/σ @deliprao]

Okay, this is crazy. TMLR editors took the unusual step of inviting authors of a subset of papers before desk rejection to have a Zoom call and answer questions about their papers. Result: only 7 of 10 invited accepted, and 7 out of 7 failed to answer any questions. People are publishing stuff they cannot explain or understand!!

https://t.co/XEsoE2UE4a

[Tim H 🦥 @creabuntur]

@binarybits For better or worse, do you think? Mixed, I guess you'd say?

[Tim H 🦥 @creabuntur]

@mattbeane @dileeplearning What some other books on this list?