Gary Marcus

Rank 14 of 47
|
Score 83
11/9/2025, 6:02:44 PM
X
In reply to:
juliette pluto 🌌
@foundjuliette
·
46d
@RomeoLupascu @GaryMarcus I hear you; and I don’t disagree on symbolic systems being far safer and more testable. That said, wouldn’t you say that humans are likewise untestable? Each individual and every new generation presents an alignment problem. and yet society functions.
Romeo Lupascu
@RomeoLupascu
·
46d
@foundjuliette @GaryMarcus There are many angles that this tech can't be constrained enough to make it safe.

When creating software (applies to hardware too) that is reliable one must ensure its total behavior is known understood and experienced at least once.

That is normally achieve in classical
juliette pluto 🌌
@foundjuliette
·
46d
@GaryMarcus Curious to understand your perspective on this; do you think it’s impossible for deep learning + RL LLMs to reach acceptable levels of robustness? Or just improbably unlikely.
Gary Marcus
@GaryMarcus
·
54d
“LLM security remains an unsolved problem”

And it always will.

We need new architectures, or we are screwed.
juliette pluto 🌌
@foundjuliette
·
72d
New paper by OpenAI, Anthropic, GDM & more, showing that LLM security remains an unsolved problem. -- We tested twelve recent jailbreak and prompt injection defenses that claimed robustness against static evals. All failed when confronted with human & LLM attackers.

The statement engages in a discussion about the safety and reliability of symbolic systems versus human behavior, in the context of technology and societal function. It touches on the challenges of ensuring safety and alignment in both systems and society.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm and contributes to a thoughtful discussion on technology and societal issues. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    It promotes understanding by comparing symbolic systems and human societal function, encouraging a deeper look at alignment challenges. [+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 constructively with previous comments, adding to the dialogue without personal attacks. [+1]