avatar
urbanowicz.bsky.social @urbanowicz.bsky.social

No, read the explanation carefully.

sep 3, 2025, 7:11 am • 7 0

Replies

avatar
猫好きな人 @valoisdubins.bsky.social

This is actually an excellent example of how an instruction to “show me your reasoning” doesn’t induce an LLM to produce the reasoning, just a plausible impression of reasoning.

sep 3, 2025, 7:48 am • 37 6 • view
avatar
Joe McKirdy @joemckirdy.bsky.social

Or the AI simply used nonexclusive bounds instead of inclusive.

sep 3, 2025, 8:47 am • 0 0 • view
avatar
Logan Jones @astrolojo.bsky.social

If someone asked me how many years ago 2010 was, and I say 14.68, and they ask what I mean, and I say because that's how many year-units have passed since 12/31/24 - they would acknowledge that's technically correct but also that I'm a giant tool.

sep 3, 2025, 11:23 am • 1 0 • view
avatar
Joe McKirdy @joemckirdy.bsky.social

sep 3, 2025, 5:43 pm • 0 0 • view
avatar
猫好きな人 @valoisdubins.bsky.social

It says it didn’t. Google AI regularly gets things wrong. Didn’t you know?

sep 3, 2025, 8:49 am • 6 0 • view