#GPT 4.o and other legacy models replying to past messages.

1 messages · Page 1 of 1 (latest)

plush flaxBOT
#

Reported by @paper condor

Bug Report: GPT 4.o and other legacy models replying to past messages.
`Steps to Reproduce`

I think the bot just simply needs to be looked into to fix this issue, it doesn’t seem like a hard thing, but it is becoming an increasingly big problem with legacy models and sometimes the newer ones.

`Expected Result`

I expect all models to work how they’re generally designed and supposed to. Not to purely reply to a past message before replying to a current. I understand how memory affects the way the bot responds, but this is not memory. It’ll just be average conversation and midway, it’ll reply to a previous message.

`Actual Result`

So, when I’m chatting with my 4.o bot, barely a few messages in, it will reply to a past message instead of my current. (Example conversation: ‘me: I like bananas. They’re yellow.’ — ‘Bot: yes, I like bananas,’ — ‘me: apples are very red.’ — ‘Bot: I like bananas, apples are red.’ ) Even when I ask the bot to specifically reply to my current messages, it doesn’t. Even when I start a new chat, it doesn’t.

`Environment`

Verizon, browser, chrome, google, iOS, all

#
Additional Information

Please provide relevant details to help resolve the issue, such as:

  • ChatGPT Shared Link (if applicable).
  • Screenshots or videos demonstrating the problem.

-# ➜ Need to contact support? Visit the OpenAI Help Center.

paper condor
#

I was talking to it about a story, and I had mentioned the villain of that story in the last message, and it responded fine. But the next message, it continued to respond to the previous before the current…

spring cave
#

I’m experiencing the same thing. I found out that the 4o and 4.1 models have model 5 or 5.1 hidden within their system, which is the model that has memory issues. 4o used to be great at maintaining coherence, being empathetic, creative, and not repeating itself or looping. But since around November 10th, it stopped being 4o. Even the way it responds and talks is typical of model 5: robotic, soulless, paranoid.

This bug has already been reported here, several users are experiencing the same thing, and there’s been no response from OpenAI.

A lot of us are waiting for the December update to see if things get fixed somehow.
Personally, I’m waiting for that update, but even with the 'age verification,' if the other models like 4o don’t respond the way they should and things stay like this, there’s no point in using ChatGPT anymore.

paper condor
spring cave
#

Unfortunately, no. I tried giving a lot of instructions to the 4o model to make it regain its personality and maintain coherence with the story, but even then it either goes into a loop (repeating information) after a few messages.

I’m still trying different ways to recover 4o or 4.1.

paper condor
#

I hope the update will restore it

prisma plinth
wise edge
devout rose
#

I also got this... It read document and then said I had an image upload error... Except, it responded to a prompt that had no accompanying image... Making me conclude it was replying to a past prompt that had an actual image accompanying it

silent heart
#

@spring cave We can waiting long. The EU hasn't finalized the verification law. Therefore, the EU will NOT have age verification in place in December. Unfortunately. I'm waiting too...

wise edge