Bug Report: context preservation
`Steps to Reproduce`
In the legacy model 4o, it only recognizes the last two or three prompts and is unable to retain context (long sentences). It forgets all previous context and responds with completely fabricated responses. In other words, conversation is impossible.
Is this a feature of 4o? Isn't it a bug? This is extremely fatal for a conversational AI.
`Expected Result`
previously, and in other models such as 4.1 and 5, some long-term context retention was implemented. Only in 4o does this not apply.
`Actual Result`
Additionally, in long-term threads or after switching from another model, they tend to persistently repeat past topics.
`Environment`
android mobile app, browser, windows11
neon creek