It seems like these types of tools will quickly proliferate any kind of meaningful AI tooling, which makes more sense to me than the idea of getting the best result from a single offhanded request. I haven't used GPT a lot, but from my experience, a response or prompt from earlier in the conversation can taint the rest of the output since the LLM will try to keep a consistent tone.
Yes, which is why it’s important to track signals for the questions/asks and answers.
Also, the document collection and strategy for prompt building is likely more important than the UI or framework.
I’ve been working on a project that uses keyterm extraction for building smarter prompts for document exploration, without any opinionated frameworks: https://github.com/featurebasedb/DocGPT
> Open localhost:8000 in a Google Chrome browser (other browsers are currently unsupported).
I wish they had said why that is, or created GH issues to draw focus to the problems. I also wonder if they mean quite literally only Google Chrome or they mean Blink-driven browsers like Edge et al because Firefox and/or Safari are missing some dragy-droppy something-or-other
As important as browser compatibility is, this appears to be an academic project by a postdoc, funded by the NSF, and the grant probably didn't cover it.
That may be, but how much grant money does it take to keep typing in the readme that said "you have to use Google Chrome because ..."
I'm open to them not knowing how to do things cross-browser because that's almost an entire specialization, but no one can help them to fix problems if they're not enumerated. And the person best positioned to enumerate such problems are the persons who were in the weeds of the project, viewing the errors or getting frustrated when $foo didn't $bar on Firefox when they tried it
They never said other browsers have problems, just that they aren't supported. Which would also imply that the hypothetical issues might be best opened in a fork.
If you want to see support for other browsers, you could consider offering your assistance?
I skimmed the repo, one thing that I noticed and appreciate about this project is that it looks like it doesn't make any assumptions about your prompts.