Fuck Claude Again

I have been using Claude a lot. And I´m eeally tired of it.

It keeps making the dumbest mistakes these days.

I really feel like they are watering it down. I’ve seen one of their engineers on HN saying that it’s not the case. So I guess I was on the lucky side of the slots machine for a while?

I have some ideas I would like to try; for example, using Groq with less competent models and the pi agent in a harness.

I don´t like the feeling of waiting for Claude to churn out crappy code that I don´t understand while reading Hacker News.

Idk. I feel like it’s like Instagram. Some people do just fine and have a healthy relationship with it, but some people can’t help but getting addicted and letting it ruin their lives.

So, yeah. I am not going to use Claude anymore. I am going to use these terrible homemade harnesses with open-source models and that’s it.

The simple fact that I am stuck with one provider is wrong.

I should be able to switch models easily for the task that I want to accomplish, switch providers hosting the model that I want.

Let’s see how it goes.