Gary Marcus

Rank 19 of 47
|
Score 64
6/23/2025, 3:43:19 AM
X
In reply to:
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.
Lawrence Chan
@justanotherlaw
·
16d
@GaryMarcus I disagree that <=0.01% of the historical effort or $ of the AIS community has gone into this -- I don't know how large you think the AIS community is, but my guess is there are still <5k FTE right now, the vast majority in the last 5 years. Let's say 30k person-years to be safe.
Lawrence Chan
@justanotherlaw
·
16d
@GaryMarcus I agree that the world as a whole is currently underallocating resources into seriously investigating alternative approaches. As you say, this is a commons problem. I also agree that the world's handling AI development is grossly negligent at best.
Gary Marcus
@GaryMarcus
·
16d
@justanotherlaw i think at most about .01% of the effort and $ has been expended looking at alternatives, if that.

if you believe that the possible consequence of betting on the wrong horse is catastrophe that kind of underallocation / premature closure is certifiably insane, especially given
Lawrence Chan
@justanotherlaw
·
16d
@GaryMarcus The problem with "make a new AI paradigm" is not a lack of desire. Many AI academics want this, especially those from an older time when making new insights about intelligence was key to AI progress.

The problem with making a new paradigm is that it turned out to be too hard.
Lawrence Chan
@justanotherlaw
·
16d
@GaryMarcus It's not that AIS people haven't considered working on 4 -- Yoshua Bengio and davidad come to mind here as contemporary efforts, Stuart Russell and Eric Drexler as older examples. And of course a lot of mech interp people in 2022-3. I think MIRI also tried this in 2018-20, but

The statement is part of a broader discussion on the allocation of resources and efforts in AI development, specifically regarding the reverse engineering of algorithms and the exploration of alternative AI paradigms. The tone is critical, suggesting that previous warnings about the ineffectiveness of certain approaches were not heeded.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not directly cause harm but implies a critique of past actions, which could be seen as constructive criticism rather than harmful.
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement does not actively promote understanding, empathy, or compassion, as it focuses on pointing out a failure rather than offering solutions or fostering dialogue. [-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 engages in constructive criticism by highlighting a perceived failure in strategy, but it does not engage in personal attacks or ad hominem arguments. [+1]