I’ve noticed the same thing at work with Opus 4.8.
ChatGPT on my personal plan does it too. Just yesterday I asked it to give some places fitting a specific criteria. The first was that they were within a 2 hour drive of my city. 75% of the locations it gave me were more than 2x that distance. It kept doing this across multiple difference searches. I tried high and pro with no difference.
That's not surprising, LLMs are bad at pulling hyperspecific facts out of memory. LLMs aren't mapping applications, they're reasoners. Just a poor problem fit
No they aren't. They're statistical token generators. They do not understand concepts such as "distance from a given location or coordinate point". If you're lucky you might ask it something likely to appear nearly verbatim in its training data, like "Chinese restaurants in Midtown Manhattan", and get back a reasonably accurate list, but it does not understand what a "Chinese restaurant" is, or what "Midtown Manhattan" is, or that one relates to the other in any way other than both appearing statistically associated with another set of tokens when they appear near each other.
ChatGPT on my personal plan does it too. Just yesterday I asked it to give some places fitting a specific criteria. The first was that they were within a 2 hour drive of my city. 75% of the locations it gave me were more than 2x that distance. It kept doing this across multiple difference searches. I tried high and pro with no difference.