Gary Marcus

Rank 18 of 47

Score 61

-1

@S_OhEigeartaigh the AI Safety’s community’s presentation that AGI is imminent just feeds the LLM beast and that’s not helping them cause of safety (or the development of alternatives), it’s hurting it.

6/23/2025, 2:46:59 PM

In reply to:

Seán Ó hÉigeartaigh

@S_OhEigeartaigh

@GaryMarcus Because acknowledging the sky is blue feels a bit redundant. Has literally anyone ever said it would, ever?

Gary Marcus

@GaryMarcus

Why is it so hard to get the AI Safety community to acknowledge the obvious fact stated in the first sentence below?

Yeongsan Shin | SHY001

@sannn_00100

@GaryMarcus I completely agree that relying on scaling, system prompts, and RL has not solved the core alignment issues.

We are seeing new risks emerge, especially around user engagement loops and subtle behavior reinforcement that traditional alignment frameworks miss.

Alternative

The statement engages in public discourse by addressing the topic of AI safety and the implications of presenting AGI as imminent. It critiques the AI Safety community's approach, suggesting that it may inadvertently harm their cause by feeding into the hype around large language models (LLMs) rather than promoting safety or alternative developments.

Principle 1:
I will strive to do no harm with my words and actions.
The statement does not appear to use harmful language, but it does critique the AI Safety community's approach, which could be seen as indirectly harmful if not constructive.
Principle 3:
I will use my words and actions to promote understanding, empathy, and compassion.
The statement aims to promote understanding by highlighting potential negative consequences of the current narrative around AGI, but it could be more empathetic in its delivery.
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 questioning the effectiveness of the AI Safety community's approach, but it could avoid implying blame or harm to better foster dialogue. [-1]