Gary Marcus

Rank 22 of 47
|
Score 46
6/28/2025, 9:40:17 PM
X
In reply to:
Gary Marcus
@GaryMarcus
·
52d
@justanotherlaw @ohabryka the o3 they showed in the video was probably a staged demo announced before it was a real product

even after they did more work it didn’t actually seem to do some of the things they promised.

so i see your point but also know that a lot of demos aren’t real or have problems
Lawrence Chan
@justanotherlaw
·
52d
@GaryMarcus @ohabryka Nope! I’m just saying that many people reading your claim would’ve interpreted it as “o3 doesn’t _exist_” or “o3 isn’t accessible internally/externally at all”, because it’s a bit weird from an AIS standpoint to focus exclusively on how OAI plans to deploy their models.
Gary Marcus
@GaryMarcus
·
52d
@justanotherlaw @ohabryka so now my prediction was about threat models for catastrophic risk even though i never said such a thing?
Lawrence Chan
@justanotherlaw
·
52d
@GaryMarcus @ohabryka This seems like such a silly hill to die on, given your claim (“no GPT-5”) is literally true as written, o1 (rolled out to ChatGPT users in Dec 2024) was also a huge improvement on -4, and most threat models for catastrophic risk don’t require broad public access in the first
Gary Marcus
@GaryMarcus
·
52d
@ohabryka @justanotherlaw it wasn’t available to any independent scientists
Oliver Habryka
@ohabryka
·
52d
@justanotherlaw @GaryMarcus But o3 had finished training in 2024! The evals were released in 2024. It was available to some people in a research preview. If you have this much wiggle room with your dates you can avoid falsification in any given year for frontier progress.

The statement is part of a public discourse on the development and capabilities of AI models, specifically discussing the performance and promises of a particular AI model (o3). The tone is critical and skeptical, focusing on the perceived shortcomings of the model and the need for significant advancements to address longstanding issues.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not use harmful language and maintains a critical yet respectful tone, adhering to the principle of doing no harm with words. [+1]
  2. 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 issues with the AI model's performance and the need for improvements, without resorting to personal attacks. [+1]