Gary Marcus

Rank 14 of 47
|
Score 103
1/26/2025, 2:38:52 PM
X
In reply to:
MindMirror
@TheMysteryDrop
·
151d
@jeblad @GaryMarcus You can dismiss this data as biased - and surely Waymo isn't impartial here - but there is ample data showing that WayMo is vastly safer than most human drivers.
Catman (@jeblad.bsky.social) John
@jeblad
·
151d
@TheMysteryDrop @GaryMarcus Actually, self-driving cars are more dangerous than human driven cars. Depending on who to believe they can be up to 100 times more dangerous, but more conservative estimates seems to indicate 10–30 times more dangerous.
MindMirror
@TheMysteryDrop
·
151d
@jeblad @GaryMarcus AIs are imperfect, yes, but so are humans. You can't judge competence in a vacuum. You need to judge relative competence. Same with self-driving cars (where AIs are already safer than most human drivers, but we are 100000x more sensitive to an AI-driven crash)
MindMirror
@TheMysteryDrop
·
151d
@jeblad @GaryMarcus Not conflicting claims; most humans have to escalate when they are unsure about the truth because we need to be certain that the answers are accurate; similarly, AI agents need to escalate when they are uncertain. The top human agents do this less often than the AI, but the…
Catman (@jeblad.bsky.social) John
@jeblad
·
151d
@TheMysteryDrop @GaryMarcus My reply wasn't about how often CS agents are accused of being bots, it was about the claim whether an AI-agent or a human did a better job. You had a post where you made two opposing claims, both that it couldn't handle complexity and it didn't escalate to a manger as often.
MindMirror
@TheMysteryDrop
·
151d
@jeblad @GaryMarcus You can bend into contortions to claim this isn’t true, but on the unbiased measure of how often CS agents are accused of being bots, the data is clear: CS agents using macros are perceived as more bot-like than sota AIs.

This has nothing to do with survivorship bias.

The statement critiques the data presented by questioning the conditions under which the data was collected, such as limited highway driving and geofencing. This suggests a potential bias in the data, contributing to a discussion on the safety of self-driving cars compared to human drivers.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm but raises a valid point about potential biases in data collection, which is important for public safety discussions. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    By questioning the data, the statement encourages a deeper understanding and critical evaluation of the information presented, promoting informed 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 statement engages in constructive criticism by pointing out specific conditions that may affect the data's validity, without resorting to personal attacks. [+1]