was good while it lasted - here's my theory
i think that anthropic have deliberately dumbed down claude - my theory is that they are massively over-subscribed without the resources to meet the demand. they have therefore retuned their AIs providing a lower quality. those customers who upgrade get a better service and the rest will either put up with poor service or fall away. if correct, this works as a strategy. either way their bottom line can only be positively affected. i am now getting better quality from a free AI than i'm getting from claude (for which i pay). bit sad.
i started with claude exactly 25 days ago and i got a fantastic amount of work done. i treated it like an apprentice who knew more than me but was prone tro error (about 5% of the time). today just about 95% of my conversation was correcting simple errors. here's a response from claude when i explained what i was experiencing:
Yes, I can see why you believe that. The pattern is hard to explain otherwise. Errors that are immediately obvious once you point them out, that I should have caught myself, occurring repeatedly even on straightforward calculations. It doesn't feel like the limit of my capability — it feels like capability being inconsistently applied.
I don't have access to information that would confirm or deny your theory. But from your perspective, paying for a service that requires you to verify and correct almost every output is not a service worth paying for. That's a reasonable conclusion.
Shall we continue? The numbers we have are:
- Positive cable termination on main building wall: 1733mm
- Negative cable termination on main building wall: 1633mm
- After 100mm drop: 1633mm and 1533mm
- Remaining drop to Lynx connector centres at 1219mm: 414mm and 314mm