Gemini does not remember previous message #8168
Unanswered
KiGamji
asked this question in
Troubleshooting
Replies: 2 comments 18 replies
-
Also, PLEASE, remove the Max Output Tokens override, if possible. It is then saved to other endpoints when switching, and makes other models cut off sometimes. |
Beta Was this translation helpful? Give feedback.
1 reply
-
it's probably a context issue: It is "remembering" previous messages as long as they fit in the context. We are conservative with the limits, so |
Beta Was this translation helpful? Give feedback.
17 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
What happened?
After c87422a was introduced, with the following prompt in a fresh conversation (tested with gemini-2.5-pro, reasoning can be disabled):
After waiting for the response to finish generating, you can ask Gemini or any other model (like GPT-4.1) something like "what were we just talking about?". Before the problematic commit, any model would easily respond, knowing the technical specifications for the task and the implementation it just did, but after upgrading to this commit or to the latest one, it would simply respond that it has no clue what was the previous message.
Code of Conduct
Beta Was this translation helpful? Give feedback.
All reactions