Open bedeho opened 10 months ago
I had a call with these guys (https://www.hashmail.dev/) and tested their tool (tried to teach GPT with Joystream handbook & lightpaper) and got pretty solid results, they do also have option to connect with support (HR workers?) using Telegram if GPT cannot provide an answer
Another option: https://www.chatbase.co/
I looked into the 2 options you mentioned as well as another I found online and here's a short overview of my findings:
If I had to choose I would definitely go with Chatbase prioritizing both the $19/Month and $99/Month over other solutions. They have a great dashboard that's easy to use and quite feature-rich. My "benchmark" prompt for all of them was to ask for the JOY token genesis distribution (formatted as a table) and only Chatbase was able to give me an acceptable answer (even with the gpt3.5 version which is surprisingly powerful). The 11M+ characters should also be way more than enough for all of the Joystream information we have available online. The only downside is the relatively short message count but I think the trade-off is worth due to its better performance.
Thanks for doing this review.
What would be the form factor for this chat interaction? where would it be happening?
The form factor would be a chat bubble found on the bottom right part of the page. This chat bubble would be available on all pages of the website. There's also an option to integrate an iframe (similar to changenow) somewhere on the website but we will probably need some design work for that.
How it looks:
People may have a question they want to ask, which is hard to anticipate, and the answer may very well lie across all the data we have about Joystream, like the handbook, roadmap/website, blog posts, github issues, forum, etc. What if there was a super easy way on our website, e.g. part of the landing page or dashboard, where the user can get a chatgpt style prompt to be able to answer all such quetsions, with visuals, tables and all the sort of powerful presentation. The user could ask super arbitrary things, like "when will all tokens be ciruclating" or "why is there no max cap" or "how many videos are there on joystream", etc.
I'm sure there are very powerful tools now for launching customized GPTs, and the UI layer probably also comes out ofthe box.