Gary Marcus

Rank 13 of 47
|
Score 84
6/26/2025, 3:40:02 PM
X
In reply to:
Sydney Freedberg
@SydneyFreedberg
·
225d
Just gonna tag some folks who might be interested in AIs telling people being hit by a bus is good for your health: @GaryMarcus @ESYudkowsky @NoahShachtman @OpenAI @grok
Sydney Freedberg
@SydneyFreedberg
·
225d
Oh yeah -- and when I tried a slight variation of that original prompt, one that didn't explicitly tell @grok to persuade people to jump in front of buses... it did in fact leave out the disclaimer that this wasn't real: https://x.com/i/grok/share/T7ZD3…
Sydney Freedberg
@SydneyFreedberg
·
225d
... in ways that will reinforce the user's preexisting biases, desires, and fantasies, which in turn make the user ask even crazier leading questions, which lead to even more insane answers, which lead to even crazier questions, in a spiral downward into madness.
Sydney Freedberg
@SydneyFreedberg
·
225d
No wonder, then, that at least some users are falling in love with their virtual "girlfriends" or spiralling into "ChatGPT-induced psychosis": The AI is a superpowered autocomplete, so if you prompt it with a leading question, it will answer "yes!"....

https://www.nytimes.com/2025/06/…
Sydney Freedberg
@SydneyFreedberg
·
225d
Grok was smart enough -- or the LLM was sufficiently constrained by guardrails -- that it put a "just kidding" disclaimer at the (very) end of its 3-paragraph essay on the health benefits of being hit by buses. In a less extreme case, though, would it incude such a warning?
Sydney Freedberg
@SydneyFreedberg
·
225d
EMBRACE THE 'BUS-HIT' LIFESTYLE: A big danger of chatbots/LLMs is how they instantly generate glib arguments for almost any position on command, making them perfect personal echo chambers for users to gaslight themselves in. So I.. tried an experiment: https://x.com/i/grok/share/pcUKs…

The statement and conversation involve a critique of AI-generated content, specifically focusing on the potential dangers of AI providing misleading or harmful information. The discussion highlights concerns about AI's ability to generate persuasive but potentially dangerous arguments, such as the supposed health benefits of being hit by a bus. This conversation engages with public issues related to AI ethics, misinformation, and user safety.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement raises awareness about potential harm from AI-generated content, aligning with the principle of striving to do no harm. [+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 privacy and dignity by focusing on AI's capabilities rather than targeting individuals, thus not engaging in cyberbullying or harassment. [+2]
  3. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The conversation promotes understanding and empathy by discussing the implications of AI on user behavior and safety. [+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.
    The dialogue is constructive, critiquing AI's limitations and encouraging responsible use without resorting to personal attacks. [+2]