Ben Millwood🔸

4619 karmaJoined

Participation
3

  • Attended an EA Global conference
  • Attended an EAGx conference
  • Attended more than three meetings with a local EA group

Comments
526

Topic contributions
1

This doesn't seem right to me because I think it's popular among those concerned with the longer term future to expect it to be populated with emulated humans, which clearly isn't a continuation of the genetic legacy of humans, so I feel pretty confident that it's something else about humanity that people want to preserve against AI. (I'm not here to defend this particular vision of the future beyond noting that people like Holden Karnofsky have written about it, so it's not exactly niche.)

You say that expecting AI to have worse goals than humans would require studying things like what the empirical observed goals of AI systems turn out to be, and similar – sure, so in the absence of having done those studies, we should delay our replacement until they can be done. And doing these studies is undermined by the fact that right now the state of our knowledge on how to reliably determine what an AI is thinking is pretty bad, and it will only get worse as they develop their abilities to strategise and lie. Solving these problems would be a major piece of what people are looking for in alignment research, and precisely the kind of thing it seems worth delaying AI progress for.

another opportunity for me to shill my LessWrong writing posing this question: Should we exclude alignment research from LLM training datasets?

I don't have a lot of time to spend on this, but this post has inspired me to take a little time to figure out whether I can propose or implement some controls (likely: making posts visible to logged-in users only) in ForumMagnum (the software underlying the EA Forum, LW, and the Alignment Forum)

edit: https://github.com/ForumMagnum/ForumMagnum/issues/10345

I agree overall but I want to add that becoming dependent on non-EA donors could put you under pressure to do more non-EA things / less EA things -- either party could pull the other towards themselves.

Keep in mind that you're not coercing them to switch their donations, just persuading them. That means you can use the fact that they were persuaded as evidence that you were on the right side of the argument. You being too convinced of your own opinion isn't a problem unless other people are also somehow too convinced of it, and I don't see why they would be.

I think that EA donors are likely to be unusual in this respect -- you're pre-selecting for people who have signed up for a culture of doing what's best even when it wasn't what they thought it was before.

I guess also I think that my arguments for animal welfare charities are at their heart EA-style arguments, so I'm getting a big boost to my likelihood of persuading someone by knowing that they're the kind of person who appreciates EA-style arguments.

Similarly if you think animal charities are 10x global health charities in effectiveness, then you think these options are equally good:

  • Move 10 EA donors from global health to animal welfare
  • Add 9 new animal welfare donors who previously weren't donating at all

To me, the first of these sounds way easier.

Thanks! (I slightly object to "the normal markdown syntax", since based on my quick reading neither John Gruber's original markdown spec nor the latest CommonMark spec nor GitHub Flavoured Markdown have footnotes)

FWIW the link to your forum post draft tells me "Sorry, you don't have access to this draft"

The onboarding delay is relevant because in the 80k case it happens twice: the 80k person has an onboarding delay, and then the people they cause to get hired have onboarding delays too.

Load more