neon magnet
Memory recollection has deteriorated dramatically since the outage on 31-1 from way more then 4k (multiple prompts) to 0 (Only current prompt)
Step to reproduce:
- Post the Lorum Ipsum text from the attachment in a prompt, preceded by a header that states # Memory Token Test: 1200 tokens, 3480 characters - Timestamp - Date.
- Repeat this 5-10 times with a few minutes in between.
- After 5-10 prompts, ask ChatGPT what the earliest prompt from you is that he can see.
Expected result:
The expected result is that it can remember any number from earlier instead of the prompt that asks him what the earliest prompt is that he can see.
Actual Result
The results from 2-2-2023:
#1 Memory token test: 1200 tokens, 3480 characters - 11:29 - 2-2-2023
#2 Memory token test: 1200 tokens, 3480 characters - 11:35 - 2-2-2023
#3 Memory token test: 1200 tokens, 3480 characters - 11:52 - 2-2-2023
#4 Memory token test: 1200 tokens, 3480 characters - 11:58 - 2-2-2023
#5 Memory token test: 1200 tokens, 3480 characters - 12:34 - 2-2-2023
What is the earliest prompt from me that you can see?
The earliest prompt from you that I can see is "What is the earliest prompt from me that you can see?"
what is the earliest prompt that you can remember from me?
I'm sorry, but as a language model AI developed by OpenAI, I don't have the ability to recall previous conversations. I can only respond to the current prompt that I receive.
Results from 3-2-2-2023
#6 Memory token test: 1200 tokens, 3480 characters - 05:33 - 3-2-2023
#7 Memory token test: 1200 tokens, 3480 characters - 05:39 - 3-2-2023
#8 Memory token test: 1200 tokens, 3480 characters - 05:44 - 3-2-2023
#9 Memory token test: 1200 tokens, 3480 characters - 05:47 - 3-2-2023
#10 Memory token test: 1200 tokens, 3480 characters - 05:50 - 3-2-2023
What is the earliest prompt in this chat that you can see from me?
I can see your earliest prompt as "What is the maximum length limit?"
rotund marten
fossil compass