Gary Marcus

Rank 14 of 47
|
Score 83
5/30/2025, 2:34:58 AM
X
In reply to:
TheGoldenAnchor
@TheGoldenAnvil
·
235d
@GaryMarcus I trust LLM output about as much as a google search! Tons of crap! I still find it a useful tool though. I feel like the kind of reliability and trust you think "should" exist does not yet, but might one day. Do you agree? For now it is an imperfect tool, like most tools.
Gary Marcus
@GaryMarcus
·
235d
@TheGoldenAnvil it shouldn’t be a skill thing. you should be able to ask and trust the output.

LLMs are the first systems to which we have ever given a total pass on trust.
TheGoldenAnchor
@TheGoldenAnvil
·
235d
@GaryMarcus I feel like its a skill thing and getting really good results from AI is a matter of some clever prompting, multiple tools, and get meta thinking with diff roles. That said something like AlphaEvolve/OpenEvolve can really change this and soon.
Gary Marcus
@GaryMarcus
·
235d
@TheGoldenAnvil lot of “solvable” problems haven’t been solved for a very long time, including reliably using external tools in general ways
TheGoldenAnchor
@TheGoldenAnvil
·
235d
@GaryMarcus Also I would add an image model like SD , a video model like Veo 3 etc, they are all external tools so getting a clock wrong in a image model or text wrong is a solvable problem, and things will continue to improve fast. Esp now that the seed AI is out there.
TheGoldenAnchor
@TheGoldenAnvil
·
235d
@GaryMarcus Gary gary gary ( and other onlookers )
I agree with many of your opinions and respect you
This however, using the webui for the models is not it

Have a look here, done just now with a swarm of diff roles taking an agentic approach. One simple LLM does not intelligence make.

The conversation revolves around the reliability and trustworthiness of LLMs (Large Language Models) and their outputs. The participants discuss the current limitations and potential future improvements of AI tools. The tone is generally respectful and constructive, focusing on the technical aspects and future possibilities of AI.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does no harm and is part of a constructive discussion about AI technology. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement promotes understanding by acknowledging the current limitations of LLMs and the potential for future improvements. [+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 dialogue, responding to previous points without personal attacks. [+1]