Gary Marcus

Rank 13 of 47
|
Score 83
4/17/2024, 1:53:07 PM
X
In reply to:
Julian Harris
@julianharris
·
634d
@GaryMarcus Closest I have seen is MSFT’s work on fact checking against Wikipedia then fine-tuning verified facts. Obviously limited to Wikipedia. https://arxiv.org/abs/2305.14292
Gary Marcus
@GaryMarcus
·
634d
That’s actually the problem here: faith.

We don’t need faith.

We need a form of AI that can sanity check its own answers.

LLMs have shown time and again that they aren’t it.
Bot man
@botyetnotbot
·
634d
@GaryMarcus LLMs are start of new gold rush. It's just a glimpse of what's possible. We will definitely see some new way to train models that will reduce hallucinations and make them more situationally aware. A lot needs to be done and it will be done by someone. I have full faith.

The statement '@julianharris And clearly not actually working in production even on wiki' is a critical response to a conversation about the application of AI in fact-checking and grounding large language models (LLMs) in reliable sources like Wikipedia. The tone is dismissive and implies skepticism about the practical implementation of the technology discussed. The content is a critique of the effectiveness of AI systems in production environments.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does no significant harm but could be seen as dismissive. [-1]
  2. Principle 2:
    I will respect the privacy and dignity of others and will not engage in cyberbullying, harassment, or hate speech.
    The statement respects privacy and does not engage in harassment or hate speech. [+1]
  3. 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. [-1]
  4. 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 borders on a personal attack rather than constructive criticism. [-1]