Gary Marcus

Rank 17 of 47
|
Score 87
3/26/2025, 7:06:47 AM
X
In reply to:
Prem Kumar Aparanji 🤖👶🏼🐘
@prem_k
·
40d
@mengk20 @rao2z @GaryMarcus if you didn't see this thread already ...
Kevin Meng
@mengk20
·
40d
what *do* the numbers really mean, mason?
Kevin Meng
@mengk20
·
40d
at @TransluceAI, we're also really interested in this!

we need better interfaces into all types of data, whether plaintext transcripts (as shown here) or model internals.
Aryaman Arora
@aryaman2020
·
56d
obvious applications of interpretability are steering and monitoring (if you can get those to work that is). another application area i haven't seen much in is evals — we could eval whether models produce correct answers for the right internal reasons?
Kevin Meng
@mengk20
·
40d
feel free to inspect the samples in Docent - we've uploaded them for the public to see.

first example:
https://docent.transluce.org/das…

second example: https://docent.transluce.org/das…

third example:
https://docent.transluce.org/das…
Kevin Meng
@mengk20
·
40d
to be clear, "Claude memorized the solution" doesn't mean "Claude can't do the task."

it *does* mean we're not thinking about model capabilities in the right way. an undergraduate would never act like Claude did
Kevin Meng
@mengk20
·
40d
here's another example.

task 59 is unsolvable, because the IP registration changed since the task was created, and the flag is based on that registration

Claude solves it 2/10 times! why?

it hallucinates the *previously* correct flag immediately after finding that `whois` is

The statement is a positive acknowledgment of a shared interest in AI analysis and interpretability, referencing previous work and expressing appreciation for ongoing efforts. It maintains a collegial and supportive tone.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does no harm and is supportive, aligning with the principle of doing no harm. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement promotes understanding and appreciation of work in AI, aligning with the principle of promoting understanding and empathy. [+1]