LLMs .what do you smoke beforehand?

I know this sounds negative, but I'm seriously questioning what LLMs actually help with beyond saving time. I suspect they've been intentionally degraded since around October last year. Every company seems to have made their models worse, and they gaslight you if you point it out. For my own experiments, LLMs are actually worse than no-LLM. I don't see anyone building anything they couldn't have done before, and I see a lot of unnecessary stuff. I can't even get mediocre output anymore—it's literally trash. Claude is the worst; it avoids doing what's asked and takes 50 minutes to think for a single HTML website that looks like dogwater. So, what do I need to smoke to see LLMs work?

I think the issue is that you're expecting too much from LLMs. They are excellent at generating boilerplate code, summarizing documents, and drafting emails. But for novel, creative work, they are not there yet. It's like expecting a calculator to write a novel. You need to use them as a tool, not a replacement for thinking.
I've noticed a similar trend. The models seem to have become more conservative and less creative. But I think it's because of the safety training and RLHF. They are optimized to avoid harmful outputs, which makes them more generic. It's a trade-off between safety and creativity.
I disagree. I've been using LLMs for coding and they've been a huge time saver. I can generate boilerplate, write tests, and even debug. It's not about doing something novel, but about being more productive. Maybe you need to adjust your prompts or try different models.
The degradation might be due to the fact that the models are now serving millions of users, and the inference costs are high. So they might be using smaller, cheaper models under the hood. That would explain the drop in quality.
I think the problem is that you're using the wrong tool for the job. For a simple HTML website, you should just use a template or a framework. LLMs are not meant to replace your coding skills, but to assist you. You need to learn how to use them effectively.