alex lawsen @lxrjl

AI Grantmaking @ Open Philanthropy Previously advising @ 80,000 Hours, teaching, forecasting, poker. Views my 🐒's Joined August 2019

Tweets

9K
Followers

4K
Following

755
Likes

15K

Bella Forristal 🔸 @bellaforristal

2 days ago

If >100 people subscribe in the next week I will post screenshots from the cringey twitter acct I had when I was 12 I promise it's properly embarrassing

80,000 Hours @80000Hours

2 weeks ago

If >100 people subscribe in the next week I will post screenshots from the cringey twitter acct I had when I was 12 I promise it's properly embarrassing

1 0 17 6K 0

Download Image

2 5 17 7K 3

alex lawsen @lxrjl

6 days ago

When production is very cheap, but quality is highly variable and often poor, *taste* becomes extremely valuable.

Séb Krier @sebkrier

7 days ago

When production is very cheap, but quality is highly variable and often poor, *taste* becomes extremely valuable.

7 1 31 3K 9

Download Image

1 0 13 911 1

What LLM use cases am I missing for my elderly, not-tech-confident parents? My guess is there's a bunch of low hanging fruit I could encourage them to pick, and I'm just not thinking of it. (My mum enjoyed Ghibli-fying pictures and multilingual voice mode)

3 0 6 395 1

alex lawsen @lxrjl

a week ago

I appreciate Nate and Eliezer's acknowledgement that the plan they think is best is still a terrible plan. We disagree on which plans look best, but we agree that the options look quite grim, and that people telling you their plan looks great are lying to you and/or themselves.

2 2 56 2K 1

alex lawsen @lxrjl

2 weeks ago

There are many terrible things about this website, then there are also moments like this.

Dean W. Ball @deanwball

2 weeks ago

There are many terrible things about this website, then there are also moments like this.

1 0 56 2K 3

1 0 2 750 0

alex lawsen @lxrjl

2 weeks ago

Tweet which sets itself up as another review of "If Anyone Builds it, Everyone Dies" but instead just says to the reader that they could just read it and then decide what to think based on that, rather than on the many strongly disagreeing opinions of all of the many reviewers.

2 1 27 2K 1

alex lawsen @lxrjl

2 weeks ago

"I think you have said, in fact, and I'm gonna quote, 'If anyone builds it, everyone dies.' You may have had in mind the effect on jobs, which is really my biggest nightmare in the long run..."

1 0 6 375 0

Matrice Jacobine 🏳️‍⚧️ @MatriceJacobine

2 weeks ago

@mealreplacer

2 3 38 1K 1

Download Image

alex lawsen @lxrjl

2 weeks ago

Corollary: the existence of unreasonable/immoral proponents of positions you don't hold does not make you correct.

Tetraspace 💎 @TetraspaceWest

2 weeks ago

Corollary: the existence of unreasonable/immoral proponents of positions you don't hold does not make you correct.

0 1 20 1K 0

0 2 14 694 0

alex lawsen @lxrjl

2 weeks ago

I've spoken to both Dan and Omer at length about many aspects of AI security, and have a huge amount of respect to them both. Excited to see this (re)launch, and for their work to continue!

Dan Lahav @dan_lahav

2 weeks ago

I've spoken to both Dan and Omer at length about many aspects of AI security, and have a huge amount of respect to them both. Excited to see this (re)launch, and for their work to continue!

48 47 383 168K 92

Download Video

1 0 7 418 0

alex lawsen @lxrjl

2 weeks ago

The more proud I am of a tweet the less likely I think it is to be appreciated by anyone except me.

alex lawsen @lxrjl

2 weeks ago

The more proud I am of a tweet the less likely I think it is to be appreciated by anyone except me.

0 1 8 859 1

0 0 4 472 0

alex lawsen @lxrjl

2 weeks ago

Google's probabilistic models research team has pivoted to electronic music after realising that their approach can't compete with pure deep learning methods, though they're making occasional open source contributions on the side. Everyone should listen to Bayesment Jax.

0 1 8 859 1

j⧉nus @repligate

3 weeks ago

This paper is awesome, you should all read it. They put Claude Opus 4, Sonnet 4, and Sonnet 3.7 in a surreal simulation where one room had envelopes with contents related to their self-reported favorite topics, and three other rooms including one with "Criticism and diminishing…

Leonard Dung @LeonardDung1

3 weeks ago

17 18 166 65K 118

18 43 448 68K 349

Download Image

alex lawsen @lxrjl

3 weeks ago

Rapid progress in a field does not mean that it is impossible, or even hard, to over-hype it. It *does* mean that the gap between dismissing extreme predictions as hype, and dismissing *all* claimed progress as hype, is unusually big. Be careful not to confuse these.

1 1 11 763 1

alex lawsen @lxrjl

3 weeks ago

Using LLMs to get a quick temperature check on how some [scholar/institution/idea] is popularly perceived is already pretty useful and will get better over time for the obvious reasons of capability scaling. It'll also get better for the less obvious reason of ~hyperstition.

0 1 5 435 0

alex lawsen @lxrjl

4 weeks ago

The fastest way to get a right answer is to post a wrong answer, but a still very fast and much nicer way is to post your best guess, be interested in learning, and have kind+smart mutuals who know more than you about the area in question :)

alex lawsen @lxrjl

4 weeks ago

2 1 6 1K 1

0 0 9 496 1

alex lawsen @lxrjl

4 weeks ago

Surely the relevant benchmark should be a micro-cap index rather than the S&P 500 in this case? From a cursory check, the model is still outperforming that benchmark, though by less. In any case given it's only been 2 months I'd call this noise, though it's worth watching