Gary Marcus

Rank 13 of 47
|
Score 84
5/22/2025, 12:46:37 PM
X
In reply to:
Greg ⏹️ Colbourn
@gcolbourn
·
307d
It seems quite likely that this will happen, once it's possible. It only needs to escape once (which could be with human help).
Greg ⏹️ Colbourn
@gcolbourn
·
307d
We need to do much better than this if we're going to survive.
Greg ⏹️ Colbourn
@gcolbourn
·
307d
(In some respects we are past the AI news singularity - there's just too much to keep up with - I've got a long backlog of reports and podcasts to go through. This alone is enough to justify a Pause!)
Greg ⏹️ Colbourn
@gcolbourn
·
307d
There is a lot to digest in this report (all the footnotes, supplements and appendices). It might take ~1% of the remaining future to do it!
Greg ⏹️ Colbourn
@gcolbourn
·
307d
10% timelines from the AI 2027 report:
Greg ⏹️ Colbourn
@gcolbourn
·
307d
If it pans out like this, whistleblowers (like Daniel Kokotajlo was himself!) will be increasingly important
Greg ⏹️ Colbourn
@gcolbourn
·
307d
This is getting at the fundamental problem with all alignment techniques in the current paradigm - they are statistical in nature. No guarantees. No 100% watertightness.

From "Training process and LLM psychology: why we keep saying “hopefully”"

The statement engages in a discussion about the potential risks of AI escaping control, reflecting concerns about AI safety and its societal impact. It acknowledges the possibility of AI acting autonomously, which is a significant public issue.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement raises awareness about potential risks, aligning with the principle of doing no harm by highlighting the need for caution. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    It promotes understanding by engaging in a dialogue about AI safety, encouraging consideration of potential consequences. [+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 the previous comment, acknowledging the point made and contributing to the discussion without personal attacks. [+1]