Gary Marcus

Rank 14 of 47
|
Score 89
3/17/2024, 12:53:38 AM
X
In reply to:
Andriy Burkov
@burkov
·
573d
@GaryMarcus Hallucinations are by design in LLMs. It's a feature, not a bug. Kind of naive to expect a feature could be fixed.
Gary Marcus
@GaryMarcus
·
573d
I would be genuinely astonished if GPT-5 is released this year and doesn’t hallucinate.

Y’all can ridicule me if that happens, and give me some props if it doesn’t, alright?
Bindu Reddy
@bindureddy
·
573d
Lots of out of control speculation about openAI’s upcoming release!

We are nowhere near continuously learning LLMs or other outlandish suggestions around AGI

I would be thrilled if GPT-5 had long context, parsed images and videos consistently, didn’t hallucinate and had…

The statement is part of a technical discussion about the limitations of Large Language Models (LLMs) in various fields. The tone is informative and critical but not disrespectful or harmful. It aims to clarify a misconception about the nature of 'hallucinations' in LLMs, suggesting that they are not a feature but a limitation due to the architecture of these models.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not intend to harm but to inform and correct a misconception. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement promotes understanding by clarifying a technical aspect of LLMs. [+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 by addressing a disagreement in a respectful manner. [+1]