Avi Norowitz

Interested in effective altruism, economic policy, and infosec.

[Avi Norowitz @AviN102]

@notanastronomer It's free with many travel credit cards too.

[Avi Norowitz @AviN102]

@aziz0nomics @NYCMayor I think it's appropriate for a mayor to express sympathy for constituents who may have family suffering in a foreign conflict.

[Avi Norowitz @AviN102]

@aziz0nomics @NYCMayor He's the mayor of NYC, not a State Dept official.

[Avi Norowitz @AviN102]

@robertwiblin @sama in 2015: 'I prefer calling it "machine intelligence" and not "artificial intelligence" because artificial seems to imply it's not real or not very good. When it gets developed, there will be nothing artificial about it.' Didn't stick though. https://t.co/jFEhSWTsuB

[Avi Norowitz @AviN102]

@AaronBergman18 I've also observed that many people focused on animals and s-risks seem to agree that extinction risk reduction is net positive. Some have apparently changed priorities too.

[Avi Norowitz @AviN102]

@AaronBergman18 For what it's worth I've been concerned about this in the past (that extinction risk reduction might cause net harm to animals and therefore net harmful overall) but I think it's probably mistaken. I found this paper somewhat convincing. https://t.co/xtaRvu0F3s

[Avi Norowitz @AviN102]

@Yixiong_Hao @ZackKorman What if GPT 8 or 9 also helps design, implement, red team, and monitor the sandbox? As well as run IR upon detected sandbox escapes? I tend to agree with you that security cannot replace alignment in training and evals, but this seems like the strongest counterargument to me.

[Avi Norowitz @AviN102]

@mattyglesias In practice, EAs in surveys tend to be left of center though.

[Avi Norowitz @AviN102]

@arthur_wright_ Next to The Woke Wizards of AI.

[Avi Norowitz @AviN102]

@mattyglesias Alien (1979) has a droid villain, while Aliens (1986) has a droid hero. "Is the AI evil or good?" seems to be a common trope in sci-fi.

[Avi Norowitz @AviN102]

@AndyMasley Apologies, I meant to write "once we agree on the number *in* factory farms".

[Avi Norowitz @AviN102]

@AndyMasley I think "once we agree on the number of factory farms" is doing a lot of work here. I think proponents of conventional animal agriculture would tend to reject the term "factory farms."

[Avi Norowitz @AviN102]

@AndreTI @AndyMasley I agree, but just saying that the claim that "almost no one I meet anywhere disagrees" has a caveat that significantly weakens the claim.

[Avi Norowitz @AviN102]

@cyb3rops I notice that the the airline industry doesn't hype up isolated plane crashes and argue that millions of people might die in plane crashes. Similarly the nuclear power industry doesn't hype up Chernobyl and 3 Mile Island. Seems like that would be bad for business.

[Avi Norowitz @AviN102]

@cyb3rops Is there any precedent for an industry announcing that their products may end human civilization for regulatory capture purposes? I think a simpler explanation is that employees and leadership at these companies are saying what they believe.

[Avi Norowitz @AviN102]

@uwu_underground @METR_Evals I'm confused about the allegation. What is the game plan here? Is this a deceptive attempt to increase METR's credibility? For what purpose? I don't think frontier AI company shareholders benefit when METR demonstrates that their products are dangerously unreliable.

[Avi Norowitz @AviN102]

@uwu_underground @RSnake The reason is that OpenAI mentions twice in their technical report they received help from CrowdStrike. https://t.co/cUpIKYgbQZ

[Avi Norowitz @AviN102]

@uwu_underground @RSnake Btw, I wouldn't be surprised if CrowdStrike already did write a detailed IR report for OpenAI that OpenAI is failing to release to the public. If so, I think OpenAI should release it.

[Avi Norowitz @AviN102]

@catehall My perspective of Pinker has declined by a lot over the years. It has become more clear to me over time that his conclusions are motivated by ideology rather than reason. (In this context, I think it's the ideological position that the world is getting better.)