Gary Marcus

Rank 19 of 47

|

Score 50

+1

@dileeplearning this conversation would become more interesting if you were more concrete and more specific

5/27/2025, 2:42:12 PM

In reply to:

Dileep George

@dileeplearning

·

19d

@GaryMarcus imo there is no fundamental solution but a series of principled mitigations.

Gary Marcus

@GaryMarcus

·

19d

@dileeplearning O4 was a step back; are there any principled solutions to hallucinations?

Dileep George

@dileeplearning

·

19d

@GaryMarcus There has been a lot of progress on reducing hallucinations. But ofc there will still be boneheaded errors.

Gary Marcus

@GaryMarcus

·

20d

@dileeplearning Depends obviously on the domain, and I have acknowledged a zillion times (if not that in that tweet) that they are useful (e.g. for coding and brainstorming), in my last book, in numerous tweets etc.

I don’t think there has been all that much progress on hallucinations;

Dileep George

@dileeplearning

·

20d

@GaryMarcus Gary, you are missing that 100% solving is not necessary for a technology being extremely useful. LLMs are in use everywhere. People are paying for it. What these problems need are mitigations, and there has been a lot of progress on that too!

Gary Marcus

@GaryMarcus

·

23d

If the leaders of the major AI companies (not mention “AI influencers” and boosters in the major media) were intellectually honest they would admit that a half of trillion dollars investment in GenAI and its infrastructure

• has not solved the alignment problem
• has not

The statement is part of a public discourse on the topic of AI and its challenges, specifically addressing the issue of hallucinations in AI models. The tone is constructive, aiming to encourage more detailed and specific contributions to the discussion.

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 suggesting that the conversation would benefit from more concrete and specific details. It does not engage in personal attacks or ad hominem arguments, maintaining a respectful tone. [+1]