Gary Marcus

Rank 19 of 47
|
Score 64
6/23/2025, 4:20:10 AM
X
In reply to:
Lawrence Chan
@justanotherlaw
·
16d
@GaryMarcus @DKokotajlo I do want to point out that more than 1% of the AIS community does agree with you that current business-as-usual alignment does not seem to be working. Maybe 25% if I had to guess.
Lawrence Chan
@justanotherlaw
·
16d
@GaryMarcus I think it's worth separating out the disagreement of 1) "will LLMs lead to AGI in <15 years" from 2) "is it feasible to align LLMs reliably".

We probably shouldn't just rehash the same conversation about 1) again. I'll instead leave that to people like @DKokotajlo, who are
Lawrence Chan
@justanotherlaw
·
16d
@GaryMarcus Yep, I agree that there have not been any stellar, compelling results, in large part because of the focus on incremental progress with current systems.
Gary Marcus
@GaryMarcus
·
16d
@justanotherlaw great. now it is less than 1%, and the 99% are focusing on hasn’t yielded any stellar, compelling results.

that should be a huge sign.
Lawrence Chan
@justanotherlaw
·
16d
@GaryMarcus Also, a lot of the efforts were done earlier on, when the community was way smaller; my guess is ~25% of total effort by AIS people in 2018 went into non-"scale the deep learning models and add some RL" approaches
Lawrence Chan
@justanotherlaw
·
16d
@GaryMarcus And of course there's the mech interp phase of 2022-3, where maybe 50% of the incoming AIS talent went into trying to "fully reverse engineer" algorithms by looking at neural networks.

The statement is part of a broader discussion on the effectiveness of current approaches in AI alignment, specifically within the AI safety (AIS) community. It questions the validity of the majority's stance and seeks clarity on their arguments, indicating engagement in public discourse about a significant societal issue.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not use harmful language and is phrased as a genuine inquiry, adhering to 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 seeks to understand differing viewpoints, which can promote understanding and empathy, aligning with this principle. [+1]
  3. Principle 4:
    I will engage in constructive criticism and dialogue with those in disagreement and will not engage in personal attacks or ad hominem arguments.
    The statement invites constructive dialogue by asking for compelling arguments from those who disagree, rather than attacking them personally. [+1]