Aaron Scher @aaronscher

"...but in the meantime there will be great companies" Joined November 2015

Tweets

722
Followers

263
Following

480
Likes

1K

Aaron Scher @aaronscher

2 days ago

This is a really high quality interview! @krystalball asked great questions, including about what I think are the best “counter arguments”.

MIRI @MIRIBerkeley

3 days ago

This is a really high quality interview! @krystalball asked great questions, including about what I think are the best “counter arguments”.

1 1 41 9K 16

0 0 3 201 0

MIRI @MIRIBerkeley

5 days ago

#7 Combined Print & E-Book Nonfiction #8 Hardcover Nonfiction

4 10 165 22K 10

Download Image

As reviews for the recent book "If Anyone Builds It, Everyone Dies" come in, I'm particularly on the lookout for compelling arguments against Chapter 11. I wrote some of my takes here: lesswrong.com/posts/P4xeb3jn…

0 0 1 95 0

Ryan Kidd @ryan_kidd44

a month ago

MATS 9.0 applications are open! Launch your career in AI alignment, governance, and security with our 12-week research program. MATS provides field-leading research mentorship, funding, Berkeley & London offices, housing, and talks/workshops with AI experts.

13 57 277 1.0M 116

Download Image

Aaron Scher @aaronscher

a month ago

I don't know what to caption this, but I find it funny research.nvidia.com/labs/adlr/file…

0 0 2 192 1

Download Image

Aaron Scher @aaronscher

a month ago

0 0 4 80 0

Download Image

Aaron Scher @aaronscher

2 months ago

My personal experience using GPT-5 so far is that it still lies / hallucinates a lot, not noticeably less than previous models. Many failures are hard to diagnose, though. I think the picture from evals is unclear too, some show less hallucination, but some it's about equal.