Gary Marcus

Rank 20 of 47
|
Score 48
1/26/2025, 1:04:10 AM
X
In reply to:
DS
@LakeDaniel11
·
147d
@GaryMarcus its insane. why can't even grok do this??? @grok and deepseek cannot tell me about what i've tweeted about...bizarre
Gary Marcus
@GaryMarcus
·
147d
Precisely: the real world is where the rubber hits the road, and I have never seen any convincing evidence that any LLM or LLM-based agent can robustly survive contact with the real-world. Ever.

And as you say, too few even test for that.
DS
@LakeDaniel11
·
147d
I notice the tech guys always test these things the same way.

with math q's./etc

but they never ask them basic research stuff like "tell me about @Nike how many times have they tweeted about baseball this year?").

And all of them fail those questions still.

cc:@GaryMarcus

The statement is part of a technical discussion about the capabilities of language models and their limitations in accessing real-time data. It critiques the inability of certain models to perform specific tasks, such as retrieving historical tweet data, and compares the expectations for different models.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm; it is a critique of technology, not individuals. [+1]
  2. 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 of language models, focusing on their limitations without personal attacks. [+1]