Will Gary Marcus dunk on OpenAI's next big model release by saying that the model still fails in predictable ways?
240
Dec 31
83%
chance

Get Ṁ600 play money
Sort by:

Fwiw, Marcus seems to be in good company saying that current architectures are insufficient for building agi. Every Smart Person™ I've heard speak on the subject agrees. But he doesn't seem to have updated his beliefs since the late 2010s when he was using the Winograd Schema Challenge as an example of something LLMs wouldn't be able to crack. Him continuing to roll out his 2019 talking points is beginning to feel like "old man yells at clouds." Metaculus has it at a 95% chance that he loses his $100k bet on AI's end-of-decade abilities. https://www.metaculus.com/questions/11199/gary-marcus-agi-bet-2030/

This market seems like free money given:

A. We'll inevitably find ways to make GPT-5 fail

B. Marcus has staked his reputation dunking on current AI and saying his cognitive frameworks are better

That is a strange way to criticize the model considering the most dangerous failures are the UNPREDICTABLE ones.