doesn’t take a rocket scientist to see why they’re saying this. all you have to do is use the various models to see within the past year they’ve all become noticeably worse for average consumer use. All the training data is out of date, all of it. The gaps between updates of said data have increased. They’re out of data to consume and it took them too long to realize that feeding off each others waste doesn’t work. combine this with the fact they’re all collectively out of money.
You’re better off just using a search engine at this point like you used to in order to get a solution. Because all these LLM’s are hallucinating now to such a massive degree in a desperate attempt to provide the user with a positive solution.
doesn’t take a rocket scientist to see why they’re saying this. all you have to do is use the various models to see within the past year they’ve all become noticeably worse for average consumer use. All the training data is out of date, all of it. The gaps between updates of said data have increased. They’re out of data to consume and it took them too long to realize that feeding off each others waste doesn’t work. combine this with the fact they’re all collectively out of money.
You’re better off just using a search engine at this point like you used to in order to get a solution. Because all these LLM’s are hallucinating now to such a massive degree in a desperate attempt to provide the user with a positive solution.
Claude insisted to me that the STM32U3 was far too new to be supported by STM32duino.
It’s been supported since March of 2025.
Hell yes the out of date training data is pissing me off. And you’re right, I would have been better off just googling it myself.
On the bright side, the weeks I spent learning STM Cube IDE may not have been a waste.
the search engines were overwhelmed with so blogspam before AI; now it’s an awful state.
But I still use them instead of AI. It’s just really hard to find a primary source. AI hallucinates. Human knowledge catalog has peaked.