I wanted to bring forward an issue in relation to the efficiency of Chat GPT in a web browser that’s been bugging me fore quite some time now. In the mobile app everything works fine.
Sometimes its obviously very practical to keep using the same chat because of the prior knowledge or rules established within it.
The problem is that chats get really slow when a chat is old and has a long history.
There are some chats I pretty much can’t use anymore because they’re just so slow (and also demanding in terms of local resources I believe).
Maybe this could be improved by simply loading less of the chat when you use it?
I understand that there needs to be some loading for context reference within the chat, but that should occur on the servers. I’m rather talking about literally loading hundreds of web messages into the users browser, which seems pretty excessive lol
I hope this can get fixed because right now im stuck making new chats or using Gemini which sucks at context because has no brain.
I haven’t experience what you describe. But I do experience the browser significantly slowing down or even crashing if what I paste in chatgpt is an extremely long text.
edit: the tag on this topic is “API” but I think you are referring to ChatGPT.
Same.
Saving some of the prompts into memory is my current-to go. However it’s still not as useful/convi as using the same lengthy chat directly. The downside of using the same chat is definitely slow/sometimes page crash. I wonder if there’s anything that could be done🤔
This seems to come up multiple times constantly here, not sure why OpenAI does not implement lazy load on the conversation history, instead they load the entire conversation history at once into the browser (this causes your browser to slow down, freeze, or hang on really long conversations). Most conversational/messaging apps lazy load the conversation, meaning the first 20 or so messages are displayed, then as you scroll up it loads more. This prevents making the browser run out of memory and hang/freeze.
Yes, I have been running into this issue as well. My workaround is manually condensing context from time to time and editing a previous prompt to now include the updated context in fewer total messages. However, this isn’t a perfect solution, and I hope that this is something that OpenAI will improve in the future.
Mobile version chatgpt works perfectly fine and fast. With the web version i have the problem where i need to refresh the site or i’ll get an even longer response since the site doesn’t respond. Also most of the times forced to start a new conversations. Kind of sucks to be paying 23 euros a month for a product with this problem.
(Optional) If you can and if not already the case, create a project with a system prompt that will give ChatGPT a lot of context to continue your discussion on the same tone, expertise, etc
(Optional) If you can and not already the case, move your long/blocked/laggy discussion inside this project
While being on the long/blocked/laggy press CTRL+A or CMD+A to copy the whole content of the page
Open Word or Google Docs, whatever
Use CTRL+SHIFT+V (and NOT CTRL+V) to paste the text without the formatting
=> You should obtain a very long Word document with the whole text dump of your blocked/laggy conversation
Remove bits of text at the beginning and end of the document that is not part of the discussion (you will see other discussions’ names, etc)
Save your Word document in a folder (Cloud-linked if possible) and name it something like mysuperdiscussion1.docx
In the ChatGPT project (if possible, otherwise just in your convos) create a new, blank discussion
Upload you .docx into the new discussion and say to ChatGPT something like “Based on this document please continue the discussion”
IMPORTANT : Remember, in time, to keep all the docx in a folder, named mysuperdiscussion1.docx, mysuperdiscussion2.docx etc
You will be able to give them all back to ChatGPT each time you start a new discussion to replace a laggy/long one. It will act as the “memory” of your conversation
