Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[BUG REPORT] It does "Summarizing..." on every single response now. #499

Open
gaveitatry2 opened this issue Oct 6, 2024 · 6 comments
Open

Comments

@gaveitatry2
Copy link

gaveitatry2 commented Oct 6, 2024

Hi. Chat GPT-Web was working perfectly for a year. But now it is acting buggy. Every time that I type something and press Enter, the chat starts "Summarizing..." before it responds. Before, it would wait around five or six responses before it started summarizing. But now it does it every single time.

Also, it is not remembering anything now. One response, a character is wearing a skirt. I type something, it summarizes, and then suddenly the same character is wearing jeans. It no longer remembers what happened the previous response.

@gaveitatry2 gaveitatry2 changed the title [BUG REPORT] It s [BUG REPORT] It "Summarizing..." every response now. Oct 6, 2024
@gaveitatry2 gaveitatry2 changed the title [BUG REPORT] It "Summarizing..." every response now. [BUG REPORT] It does "Summarizing..." on every single response now. Oct 6, 2024
@Niek
Copy link
Owner

Niek commented Oct 7, 2024

Strange, I cannot reproduce this. Are you using https://niek.github.io/chatgpt-web/? Care to share any screenshots including devtools output?

@gaveitatry2
Copy link
Author

gaveitatry2 commented Oct 7, 2024

Strange, I cannot reproduce this. Are you using https://niek.github.io/chatgpt-web/? Care to share any screenshots including devtools output?

Yes. That is the website that I am using and have used for the past year. Thank you for the website, by the way. It is terrific. The best thing that ever existed. I'm reluctant to take screenshots at the moment, but if it comes to that, I will see what I can do. I don't even know what a devtools output is. Just now, I tried to reproduce this in another browser, and I had the same problem. Basically, I am using the Jen profile with the StableBeluga-2-70b model. After the System Prompt, I type something, and it responds without Summarizing. But after that initial response, if I type anything else, it immediately starts Summarizing before responding. And it will keep doing that each and every response from then on. Always Summarizing before responding.

@gaveitatry2
Copy link
Author

Strange, I cannot reproduce this. Are you using https://niek.github.io/chatgpt-web/? Care to share any screenshots including devtools output?

Okay, so previously, for the past year, I would need to type in the Chat GPT-Web around 6-10 times and get responses 6-10 times before the AI does a summary. But now it is doing a summary each and every response.

Also, previously, for the past year, the AI always paid attention to previous responses. So if a character was wearing a white t-shirt and blue jeans in the previous response, the character would still be wearing a white t-shirt and blue jeans in the next response. But now the AI has a very bad memory. The character could be earing a white t-shirt and blue jeans in the previous response and a gray sweater and a dress in the next response.

I have not changed my settings in the past year. I am using the same settings as before. That is why this is strange.

The following are two videos that shows what is going on.

Summarizes Every Response Now And Does Not Remember Previous Responses
https://youtu.be/ffQiDbe4aTA

Chat Settings
https://youtu.be/86xfRY2ft_o

@Niek
Copy link
Owner

Niek commented Oct 8, 2024

I think this is due to the low token threshold you have in your chat settings. Maybe you can increase this for this particular model?

@gaveitatry2
Copy link
Author

I think this is due to the low token threshold you have in your chat settings. Maybe you can increase this for this particular model?

The default Token Threshold for the Jen profile is "0.8". But for the past year, I have had it set to "4000" because I ran into some problems and someone recommended setting it to a value close to but less than "4096". So isn't it already pretty high? Also, this setting hasn't changed for the last year. I have the same settings now that I have been using for the past 12-14 months. But this problem I am having with the Summarizing every response and not remembering info in previous responses only happened sometime in the last month. Or are you talking about another setting? Could you screenshot which setting you want me to change and what value I should change it to?

@gaveitatry2
Copy link
Author

I don't understand why it is Summarizing.... every single time now. It is ridiculous. It used to wait six to ten messages before it did one summary. But now it takes way longer to generate a response because it has to summarize first. And now when I scroll up, I see summary after summary... just a long list of summaries. Whereas before, I would scroll up and see like six of my messages and six of the AI's messages, before I see one summary, then scroll up and see another ten messages before seeing another summary. And the summaries are weird now. They don't have any useful information and actually include information that shouldn't be in there. And the AI is no longer paying attention to the content of previous responses. It feels like someone made some changes to the code or something sometime in the last 1-3 months and it messed everything up. If that is the case, I wish that there was a rollback to make it back to how it was before. Like I said before, I'm using the same settings that I have used for the past 14 months. So it isn't my settings and it isn't some new setting that was added that I didn't notice or pay attention to. It's the behavior that has changed. Any chance for a rollback?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants