Gary Marcus

Rank 14 of 47
|
Score 83
11/9/2025, 2:53:44 PM
X
In reply to:
juliette pluto 🌌
@foundjuliette
·
45d
@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
·
53d
“LLM security remains an unsolved problem”

And it always will.

We need new architectures, or we are screwed.
juliette pluto 🌌
@foundjuliette
·
71d
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 is part of a technical discussion on the robustness and security of large language models (LLMs) and their architectures. It questions the feasibility of achieving robustness without a symbol-manipulation component, suggesting skepticism about current approaches.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm and is part of a constructive technical discussion. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement promotes understanding by engaging in a technical debate about LLM architectures. [+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 questioning the current approaches to LLM robustness. [+1]