Untitled topic • Hacker News Board

#1

Post by tranchms » Mon, Jan 20, 2025, 2:36 AM UTC

No post body was provided.

Re: undefined

#2

Post by zoogeny » Mon, Jan 20, 2025, 2:41 AM UTC

Hotz claims that Axios is reporting this but the only relevant article on Axios I could find [1] did not mention the cracking crypto part (or anything about polynomial time algorithms for factoring).

Big if true, but until we get some kind of official confirmation I will remain skeptical. I wouldn't put it past Hotz for this to be a troll for the lols.

1. https://www.axios.com/2025/01/19/ai-superagent-openai-meta

Re: undefined

#3

Post by iab » Mon, Jan 20, 2025, 2:49 AM UTC

zoogeny wrote: post #2

Hotz claims that Axios is reporting this but the only relevant article on Axios I could find [1] did not mention the cracking crypto part (or anything about polynomial time algorithms for factoring). Big if true, but until we get some kind of official confirmation I will remain skeptical. I wouldn't put it past Hotz for this to be a troll for the lols. 1. https://www.axios.com/2025/01/19/ai-superagent-openai-meta

This is another Hotz troll-job

Re: undefined

#4

Post by A_D_E_P_T » Mon, Jan 20, 2025, 3:15 AM UTC

The claim appears to be false.

But it was likely motivated by a real phenomenon: Feed it the right prompts, and O1 will often claim to solve various open questions in mathematics and computer science. Sometimes the suggestions are actually real and viable -- e.g. an iterative program on the Weaire–Phelan structure to find better solutions to the 3D tiling problem -- but usually there's a flaw in its reasoning somewhere. Worst case, it'll just make shit up. "Y-yeah, I discovered an algorithm that can crack crypto in real time" definitely falls into that category.