Gary Marcus

Rank 14 of 47
|
Score 83
6/21/2025, 4:44:22 PM
X
In reply to:
Rob Miles (in SF)
@robertskmiles
·
212d
@GaryMarcus @kevinroose A *perfect* next token predictor is a wildly superintelligent AGI. LLMs of course aren't perfect next token predictors, and can't be, and they have serious and possibly very limiting shortcomings. But this is not because they're auto complete
Rob Miles (in SF)
@robertskmiles
·
212d
@GaryMarcus @kevinroose I can stream 4K video and simulate huge photorealistic worlds in real time, saying it's just a calculator is missing the point.
There's nothing fundamental about the token prediction task that puts any limit on the capabilities of systems that perform it
Rob Miles (in SF)
@robertskmiles
·
212d
@GaryMarcus @kevinroose I don't think the claim is that LLMs aren't fundamentally doing ~the same task as autocomplete, but rather that this doesn't say much. Like how a modern PC is "just a fancy calculator". Fundamentally yeah, but it can do a ton of important and useful things a calculator can't
Gary Marcus
@GaryMarcus
·
212d
@kevinroose you honestly don’t appear to know what you are talking about. you can’t repeat multiple passages of Harry Potter verbatim without some massive autocomplete abilities and auto completion is literally the training objective.

you are embarassing yourself.
Kevin Roose
@kevinroose
·
212d
I used to think the "fancy autocomplete" stuff was just goofy cope from NLP dead-enders, but maybe it's actually more calculated than that. There's clearly a market for people who want to be told that AI is a fad and that they don't need to worry about it.
Nathan Benaich
@nathanbenaich
·
213d
it’s truly nonsensical to say things like this when genai is legit super useful for consumers and businesses - i mean, i get the academic shtick, but it is automating a ton of work already

get over it :)

The statement critiques a thought experiment about AI, suggesting it lacks engagement with the concept of generalization. The tone is analytical and seeks to clarify a misunderstanding in the discussion about AI capabilities.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm and aims to clarify a point, 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.
    It promotes understanding by addressing a specific aspect of the AI debate, encouraging deeper engagement with the topic. [+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 without personal attacks, focusing on the content of the argument. [+1]