grim narwhal
The text input shouldn't just vanish when there's an error - for long and complex stuff, it can be a hassle to put text into the text input - only to then have it be deleted.
Also, as far as browing with Bing goes... Can't you default to Bing's internet cache first? Especially for things like articles that aren't likely to change much?
It has a very high failure rate for being able to click on a link and actually managing to read it - maybe because of anti-robot systems on those sites.
So, at a minimum, it should default to Bing's cache if it can't read a page.
Currently, I'm having to do awkward stuff like break up pages from the internet into chunks so that it ChatGPT 4 can read it ... And each of those chunks of content I give it ends up counting towards by 25 messages / 3 hours.
It's a sad state of affairs.
And what is this "This conversation is too long, please start a new one" nonense?
I thought you were saving embeddings and summaries so that the chat could at least kinda go on indefinitely - even if some information might be lost.
How about enabling the 32K token model for conversations that end up long? (Before people have to start looking at Anthropic as a backup) - I don't like Claude vs ChatGPT...