Experimenting with other ways to use the OpenAI API, and one thing that has struck me is the price difference between GPT 3.5 and GPT 4. It is dramatic. If it wasn’t for the token limit of 3.5, I’d just use that for everything. It’s still good.

@manton Can’t make it so that users can plug in their own API key? That may sound technical but a detailed how-to might help.

@pratik In theory, yes, but I much prefer rolling the cost into Micro.blog Premium, like we do for podcast transcription. That way the user doesn’t have to think about it, and we just eat the extra 25 cents or whatever it might be. Just need to be careful with GPT 4 because it can add up.

@manton Really makes the $20/month for GPT Plus feel like a pretty solid value

@manton, Ideally, yes, and I would love it if you do, but if the (real) fear of running up the cost is keeping you from implementing a cool feature, then it’s the next best thing.

@aaronschendel Yeah, $20/month for Plus is a fair price. It really is way more powerful.

@jhull Interesting, I hadn’t thought about chaining. My current experiment is summarizing web pages, so I pass the whole text to it. I use GPT 3.5 if it’s short enough, otherwise GPT 4. Not sure if it would be as good if split apart.

@manton I do it all the time with screenplays. Summarize scenes, then acts, then use gpt-4 to analyze. Any summary/meta stuff I run through 3, heavy reasoning “thinking” goes through 4. Works quite well.