Gary Marcus

Rank 15 of 47
|
Score 76
Gary Marcus
@GaryMarcus
·
267d
Come for the vowels! Stay for the bicycle!

Grok 3, like all other LLMs before it, is hit or miss. Here are four misses that I easily found in my first 20 minutes of experimenting.
2/18/2025, 11:36:53 PM
X
In reply to:
Alex Volkov (Thursd/AI) 🔜 AIENG summit NY
@altryne
·
267d
@GaryMarcus @karpathy I literally just tried this again, with 6 retries, it didn't get the numbers only 1 time

You're doubling down on argument from authority, when you in fact, did not check anything yourself, and when you tried, you checked an irrelevant model.

I don't apologize when there's…
Gary Marcus
@GaryMarcus
·
267d
@altryne @karpathy and yes your original post smugly implied that i had made up my claim; i did not. so you were in fact smug, rude and factually incorrect.

the other guy apologized; you don’t seem to get it.
Alex Volkov (Thursd/AI) 🔜 AIENG summit NY
@altryne
·
267d
@GaryMarcus @karpathy This example is not at all similar, they have an auto regressive image model connected to Grok3, which nobody said was updated together with Grok the LLM. It's being used as a tool call.

Regarding "coming across as"... even grok himself doesn't think I said anything close to…
Gary Marcus
@GaryMarcus
·
267d
@altryne @karpathy 1. i got my fact in this case from Andrej; he is a legit source. At that point I didn’t know how to access without premium+.

2. Now that I figured how to access without additional cost I just tried a similar example that was not publicized and got this result below, and see the…
Gary Marcus
@GaryMarcus
·
267d
Grok 3, “smartest AI on earth”
Alex Volkov (Thursd/AI) 🔜 AIENG summit NY
@altryne
·
267d
@GaryMarcus @karpathy I don't believe I ever was rude to you Gary, but definitely not in this instance.

You in fact did not check your facts, your source was Andrej, and I'm not saying he's wrong, but replicating this on your own (like me and Ivan did) would be considered "checking your facts"

I…
Gary Marcus
@GaryMarcus
·
267d
Incredible how this guy smugly thinks I didn’t check my facts, when of course I had a source.

In fact I got my report from @Karpathy (screen capture below), and the fact that Grok 3 *sometimes* can count the letter l’s in a specific word, especially after a massively publicized…

The conversation involves a debate about the accuracy of AI models, with participants challenging each other's claims and evidence. The tone is defensive and somewhat confrontational, with accusations of rudeness and factual inaccuracies.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not strive to do no harm, as it includes accusations and a defensive tone, which could escalate the conflict. [-1]
  2. Principle 2:
    I will respect the privacy and dignity of others and will not engage in cyberbullying, harassment, or hate speech.
    The statement does not respect the dignity of others, as it involves personal attacks and accusations of unfairness. [-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 does not engage in constructive criticism, as it focuses on personal grievances rather than addressing the issue constructively. [-1]