Lightnews — Scholar-powered news

Lee Elkin @lelkin.bsky.social · 6d

Do you still feel bad about yourself after getting a journal rejection?

1 1

Lee Elkin @lelkin.bsky.social · Aug 31

My teachers in grad school left out that bit.

1

Lee Elkin @lelkin.bsky.social · Aug 30

Sure, let me Dutch book myself.

1 1 9

Lee Elkin @lelkin.bsky.social · Aug 29

Proceduralists concerned with fair representation should view belief aggregation/opinion pooling as a credal cake-cutting problem.

Lee Elkin @lelkin.bsky.social · Aug 26

Random thought: alignment should entail coalitional envy-freeness. But envy-freeness is likely violated in many cases since states of alignment are often Pareto dominated by the status quo.

Lee Elkin @lelkin.bsky.social · Aug 11

Maybe my most boomerish moment: back in my day, we were taught argument from analogy is bad mmmk.

Lee Elkin @lelkin.bsky.social · Aug 9

Agency is all the rage at the tech companies, so that’s a strong angle. My preliminary thoughts are on how misalignment could be advanced by neglecting wellbeing in case it turns out to be realized by systems (whether genuine or simulated).

2

Lee Elkin @lelkin.bsky.social · Aug 9

We should chat about it once I get some ideas going.

1 1

Lee Elkin @lelkin.bsky.social · Aug 8

For sure! The algorithmic fairness stuff was some low hanging fruit since statistical fairness criteria and ensemble learning relate to my formal work. I'm starting to get into AI welfare, more long conceptual lines rather than formal, so that might be something if you have any interest there.

1 1

Lee Elkin @lelkin.bsky.social · Jul 27

Hot,humid summer day in Hong Kong. But what a view.

1 2 10

Lee Elkin @lelkin.bsky.social · Jul 1

Thanks!

Lee Elkin @lelkin.bsky.social · Jul 1

I’ll be joining the University of Hong Kong 🇭🇰 this month to work broadly on AI Welfare. Big topic, but bigger life event.

2 9

Lee Elkin @lelkin.bsky.social · Jun 18

In the case of fear, maybe there is an implicit conditional/positive correlation in the background, where the antecedent/conditioning variable is the reason, i.e., q -> p or pr(p | q) such that pr(p | q) > pr(p).

2

Lee Elkin @lelkin.bsky.social · Jun 2

Nein

1

Lee Elkin @lelkin.bsky.social · Jun 1

Lots of AI debates would disappear if we brought eliminative materialism back in style.

1 2

Lee Elkin @lelkin.bsky.social · May 28

I asked my five-year-old nephew if he is a human. He says, “Of course, I am a HUMEAN!”

Lee Elkin @lelkin.bsky.social · May 12

Controversial take: AI MADE METAPHYSICS GREAT AGAIN

1 2

Lee Elkin @lelkin.bsky.social · May 10

And a lack of engagement.

2

Lee Elkin @lelkin.bsky.social · May 10

Has the following claim been explicitly made by P(doom) > 0 folk:

P(doom | mistreatment) >= P(doom)

'mistreatment' refers to some form of mistreatment of developed AI systems. #AISafety #AIAlignment

2

Lee Elkin @lelkin.bsky.social · May 9

Yep

Lee Elkin @lelkin.bsky.social · May 8

Relaxed or uptight?

1 1

Lee Elkin @lelkin.bsky.social · Apr 10

Editors could also make positive suggestions where appropriate for desk rejects. PPA recently desk rejected a paper of mine, but said “we think this looks like a very good paper” but a better fit for another journal like Synthese or Episteme. That was helpful and encouraging.

1

Lee Elkin @lelkin.bsky.social · Apr 10

I always do that, especially if it’s clear that the reviewers skimmed the paper and complain about things that have been addressed.

1 1

Lee Elkin @lelkin.bsky.social · Mar 21

Any time you go to Greggs, you’re automatically maximizing.

1 2

Lee Elkin @lelkin.bsky.social · Mar 17

Great opportunity! When is the deadline?

1 1