You have to inform it in a way it can process that it needs to consider clicking links to be piloted by a human so it can send a web browser ID.
Try the OMARGOS ChatGPT Browsing edition from the -OMARG-AIR-AID repository on github’s models folder. Reach out if you need help setting it up.
Is it confirmed that the Web Browsing plugin cannot read dynamic websites and only requests the source code?
Most annoying thing for me is that “browsing” can browse github repo index page, but when trying to get contents: click failed
web pilot’s prompt is fully controllable by ChatGPT, you just have to let it know it can do it.
It can use APIs once it’s aware that it can. It can do anything a browser can, so long as you teach it that it IS a web browser and not a bot when browsing under user direction.
It’s possible that browsing can also do stuff like this, but without any info on the actual back end there’s no telling if it can.
However it appears it’s been replaced by Bing in some way, and it does seem to perform a lot better now when directing it to explore repositories designed to teach it web browsing tricks.
@kiki51669 @ThioJoe @elmstedt @DarkIlluminatus So my plugin “Web Requests” just went live yesterday, and here’s an example of what it can do:
More or less, plus, we just dont save data. All in memory caching for performance and pagination. Don’t even have a database. And wont.