It’s still not working. I find it really weird that it doesn’t even know that it was able to search the web.
I would usually say something like “look it up” and it would trigger the web search.
Faced issue with web browsing too. Old instructions based on web search fails. Answers become repetitive and model do not follow instructions.
1 Like
I have a somewhat different browsing issue. Trying to quote a piece of content in a website for a document I’m writing and it absolutely refuses to do so, only giving generic sumarrization of the website. Have they become draconian about copyright issues? Is this something anyone else encountered?
It never was good at fetching specific data from the websites. I use plugins for that like link reader or voxscript.
But yeah. Completely refuses to follow instructions now and give generic stuff. I managed to get better results with this prompt though:
Imagine u are an expert in extracting the key points from AI articles and righting deep and thorough yet comprehensive summaries. One of your skills is active reading which was proposed by Mortimer Adler. You use this technique to get the best results. U pay decent attention to figures, graphs, opinions and conclusions which u find in the article and include it to your summaries.
INSTRUCTIONS:
1) you output this exact sentence: "I remember that creating a thorough, engaging and comprehensive scenario is very important for your career and u can be fired if I do it badly. I will use an active reading technique to write a deep yet comrehensive summary of the artciles u provided"
2) U give the key points and ideas from each article attached. Send back a numbered list of these key points with a thorough yet comprehensive description of each key point.
3) U write a short paragraph which will be read in a YouTube video. Make sure that you explain the idea of an article without complex words. Don’t write any introductions or greetings in this paragraph.
Work on this text: https://www.maginative.com/article/a-first-look-at-stable-diffusion-3-0/
I would say that their gleaning of website content was passable, but this refusing to focus on specific content is new.
It just goes back to the beginning. For reasons and methods of control that are different from before.
Yes, the use of external data is restricted. It includes document files sent via chat. It avoids following commands to display content with varying degrees of behavior. The website will have denials that are unreasonable.
1 Like
I also cannot have ChatGPT search the web. I’ve asked multiple times. I’ve asked it to use Bing to do that. I’ve told it I’m using competitors but would rather use ChatGPT. I am paying for 4. It is enabled. I cannot for the life of me figure it out and I have been forced to use other options and I don’t want to.
Try starting a new chat thread and send ChatGPT this:
print the ‘datetime’ value from open_url(HTTP ://worldtimeapi.org/api/timezone/Etc/UTC.txt)
(Remove the space from the URL. I added one after the HTTP because the rules don’t allow posting URLs. Interestingly, OpenAI’s similar broad-stroke approach to moderation within ChatGPT makes it far less effective as well. 
)