Gary Marcus

Rank 19 of 47
|
Score 64
6/8/2025, 3:27:16 PM
X
In reply to:
Carlos E. Perez
@IntuitMachine
·
32d
@GaryMarcus @AvpElk Agentic AI rests heavily on the ability of LLMs to select the right tools to solve a problem. IDE's like cursor are dependent on reliable execution by the tools underneath. However a lot of the reasoning is knowing which tools to employ to solve a problem (and also a lot of
Gary Marcus
@GaryMarcus
·
32d
@AvpElk @IntuitMachine 1. yes neurosymbolic is the way to go
2. but can you reliably get LRMs to invoke the tools when they need to?
AVP
@AvpElk
·
32d
@IntuitMachine @GaryMarcus For a simple example, you don’t have an LRM multiply two 50-digit numbers using reasoning tokens. You have the LRM use a calculator to multiply the numbers.
AVP
@AvpElk
·
32d
@IntuitMachine @GaryMarcus I could see how it might be true that an LRM “𝙘𝙖𝙣 𝙣𝙚𝙫𝙚𝙧 𝙗𝙚 𝙧𝙚𝙡𝙞𝙚𝙙 𝙤𝙣 𝙩𝙤 𝙥𝙧𝙤𝙥𝙚𝙧𝙡𝙮 𝙚𝙭𝙚𝙘𝙪𝙩𝙚 𝙖𝙡𝙜𝙤𝙧𝙞𝙩𝙝𝙢𝙨.”

But that’s arguably also true with humans. The solution would be to have the LRM write algorithms that can be run using non-AI.
Carlos E. Perez
@IntuitMachine
·
32d
@GaryMarcus I disagree with its interpretation and yours.
Carlos E. Perez
@IntuitMachine
·
32d
Looking at the "Illusion of Thinking" through the lens of the QPT meta-cognitive framework reveals a profound weakness in Large Reasoning Models' epistemic self-awareness - their ability to know what they don't know and calibrate their cognitive efforts accordingly.

LRM
Gary Marcus
@GaryMarcus
·
32d
𝗔𝘀𝘁𝗼𝗻𝗶𝘀𝗵𝗶𝗻𝗴𝗹𝘆 𝗳𝗲𝘄 𝗽𝗲𝗼𝗽𝗹𝗲 𝘀𝗲𝗲𝗺𝘀 𝘁𝗼 𝗵𝗮𝘃𝗲 𝗮𝗽𝗽𝗿𝗲𝗰𝗶𝗮𝘁𝗲𝗱 𝘁𝗵𝗲 𝘁𝗿𝘂𝗲 𝘀𝗶𝗴𝗻𝗶𝗳𝗶𝗰𝗮𝗻𝗰𝗲 𝗼𝗳 𝘁𝗵𝗲 𝗻𝗲𝘄 𝗔𝗽𝗽𝗹𝗲 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗽𝗮𝗽𝗲𝗿.

I will break it down for you, in four sentences:

• It is not just that these

The statement is part of a technical discussion about the capabilities of Large Language Models (LLMs) and their ability to select appropriate tools for problem-solving. It contributes to a broader conversation on AI and reasoning models, focusing on the importance of tool selection in AI problem-solving.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does no harm and is a neutral contribution to the discussion.
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    It promotes understanding by agreeing with the importance of tool selection in AI. [+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.
    Engages constructively by adding to the conversation without personal attacks. [+1]