Is it me, or in the last week has every llm model gotten stupider?
I had a flow going, and was intentionally putting off the Devin update. The day finally came, and I had to restart my computer. I still hadn't hit the upgrade button, but when I opened windsurf for work it had transformed. No worries.. I took it in stride, hopped through some hoops, and went to work, but somehow all of my training, rules, and protocol had not only been forgotten, I have to keep reminding the AI regardless of rules, model updated knowledge base, and RAG — Retrieval-Augmented Generation protocol.. Many times I find myself having to re-word, and re-prompt several times to get a right response even though outside of Devin using the very same version, and model I immediately got my answers, and alignment. Devin desktop forgets between prompts, ignores rules, and just plain screws up all over the place while (happily) burning through tokens...
So unless anyone has a real timeline, and solution, I can't help but feel the windsurf sellers are just singing with glee... 'We got your money... We got your money!!' While we just suffer.
Get it together people. Even if you have to roll back. I think many are being patient, but with so many other options we're going to need a solid plan, commitment, disclosure, times, and dates. I for one will check back occasionally, but am taking the bulk of my work elsewhere for now, because it feels like trying to work it out with a disloyal spouse giving only 60%. So... Nope!