Gary Marcus

Rank 31 of 47
|
Score 0
4/22/2024, 11:22:59 PM
X
In reply to:
Laura Edelson
@LauraEdelson2
·
323d
@GaryMarcus I think we don’t fully understand the security risk profile of large models, and how much vulnerability is in weights vs. code. History suggests that more open is better, of course! My intuition is that most risk is application-specific, which means code.
Gary Marcus
@GaryMarcus
·
323d
@LauraEdelson2 Does this apply to open-weight models that are not truly open source?

I would also be interested in your arguments re bad actors and potential future bioweapons that you dismiss.
Laura Edelson
@LauraEdelson2
·
323d
Open source has also won the argument about security. Everyone (good and bad) finds bugs faster. But on balance, this appear to make software more secure, because there are more defenders than attackers.11/12
Laura Edelson
@LauraEdelson2
·
323d
And even outside of models, open source software has won the argument. Open source means that everyone can innovate faster, so everyone counts on making more money because everyone gets more out of the common ecosystem than they put in.10/12
Laura Edelson
@LauraEdelson2
·
323d
That's why DataBricks debuted (open) DBRX before they IPO'ed - it's critical to their business success because they make money on ALL THE STUFF that gets built on and around a model (especially by them.)9/12
Laura Edelson
@LauraEdelson2
·
323d
So why haven't all these OS/OW models stopped other models in their tracks? Even upstarts like Mistral or DataBricks? Because the foundational LLM just IS NOT the thing that gets sold, and never has been. What all these companies sell is helping you use the LLM.8/12

The statement engages in a substantive discussion about the risks associated with open-sourcing large language models (LLMs). It highlights various immediate risks such as disinformation, cybersecurity, bias, and environmental costs, among others. The tone is cautious and informative, aiming to contribute to a broader understanding of the potential negative impacts of generative AI technologies.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement aims to do no harm by raising awareness of potential risks, thus contributing positively to harm reduction in AI development. [+2]
  2. Principle 2:
    I will respect the privacy and dignity of others and will not engage in cyberbullying, harassment, or hate speech.
    It respects the dignity of others by not targeting any specific group but rather discussing the technology's impact broadly, promoting a respectful discourse. [+1]
  3. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement uses its platform to promote understanding of a complex issue, enhancing public knowledge and awareness. [+2]
  4. 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.
    It engages in constructive criticism of AI development practices, fostering a dialogue about how to manage and mitigate risks. [+2]
  5. Principle 6:
    I will use my influence for the betterment of society.
    The user uses their influence to educate and inform the public about critical issues in AI, aligning with the principle of using influence for societal betterment. [+2]
  6. Principle 7:
    I will uphold the principles of free speech and use my platform responsibly and with integrity.
    The statement upholds the principles of free speech responsibly, contributing to an informed discussion without spreading misinformation. [+1]