Gary Marcus

Rank 16 of 47
|
Score 76
6/18/2025, 2:41:59 PM
X
In reply to:
The Galois Connection
@TheGaloisCxn
·
1d
@GaryMarcus @Rahll This shouldn't be hard for people to accept given how image models like Imagen will produce near perfect reproductions of copyrighted characters from a plain text prompt.

For example, Imagen has clearly memorized Darth Vader's likeness.

This was generated completely from text:
The Galois Connection
@TheGaloisCxn
·
3d
So, I've accidentally created a whole new kind of #AI image fusion algorithm apparently.

It works nothing like conventional style transfer or Midjourney's blend feature.

I swear this isn't bad Photoshop, it's AI compositing.
Gary Marcus
@GaryMarcus
·
1d
Please stop bitching about the term “stochastic parrots” (coined by Bender et al).

No metaphor is perfect, but this new study (and my 2024 article with @rahll that the Disney lawsuit cites many times) show decisively that both sides of the metaphor are correct.

Llama can
Gary Marcus
@GaryMarcus
·
2d
Geoff Hinton. 2023: “Gary Marcus .. is making up .. how he thinks it works. [LLMs] don't pastiche together text that they've read on the web, because they're not storing any text.”

New study: “Meta's AI tool Llama 'almost entirely' memorized Harry Potter book… Llama 3.1 has

The conversation revolves around the capabilities and implications of AI models, particularly in terms of memorization and reproduction of copyrighted material. It touches on issues of copyright, AI ethics, and the potential legal implications for AI companies.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The discussion raises awareness about potential harm caused by AI models memorizing copyrighted content, aligning with the principle of striving to do no harm. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The conversation promotes understanding of AI capabilities and limitations, fostering a more informed public discourse. [+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 dialogue engages in constructive criticism of AI models and their implications, avoiding personal attacks and focusing on the issues. [+1]