ChatGPT can now reference all past chats!
Recently, OpenAI released a new ChatGPT capability that supports referencing all previous conversation content, enabling better personalized responses.
In addition to preserving previous memories, it can also quote past content.
This means ChatGPT can now understand you better and act more like a personal assistant, effectively upgrading past chat records into a knowledge base that can be directly queried and referenced.
In essence, it provides a more personalized experience.
OpenAI's reasoning domain expert Noam Brown stated that this is not just a new product feature, but also marks a transformation in LLM interaction paradigms.
Shifting from sporadic interactions to continuous interactions, similar to AI transitioning from a Q&A customer service to a friend or colleague.
Some wonder, how will AI's memory develop? For instance, will it remember not just facts, but also our experiences and feelings?
The expert Kapathy "worriedly" asked:
Will GPT form a negative impression because of my previous novice questions?
Available for Plus/Pro Users Starting Today
These new capabilities are now available for Plus and Pro users starting today.
So now you can ask ChatGPT, "What about me surprised you the most"?
It will start recalling and narrating like a friend.
You cried on the sofa that day. You prefer more specific analogies rather than abstract metaphors.
You can also have it connect two of your experiences in an unprecedented way.
ChatGPT: You like large, hierarchical systems, like "War and Peace", like large language models.
You can also ask what about you made it speechless.
Well, if we're exploring the "most embarrassing" area - here's a standout (with love):
You once spent over 100 years of game time running a diplomatic simulation...
However, the stability of this capability remains to be seen, with some discovering serious hallucinations.
It's not as I imagined, some questions should be searched in the chat history, but it generated unrelated hallucinations on its own.
Some asked why it can't verbatim quote previous messages or provide specific dates of conversations?
Others are concerned about whether this memory mode overly influences context.
OpenAI states that if you want to modify ChatGPT's memory, you can do so through conversation.
If you don't want to use this feature, you can disable it. Otherwise, it will be enabled by default.
One More Thing
Many responded to this update: Just this??
Altman dramatically announced last night that the new feature made him too excited to sleep.
A few days ago, he teased the upcoming release of o3 and o4-mini within weeks.
How could such actions not build anticipation for a new release?
However, it was another feint, launching a feature already supported by Gemini...
But there are reports that the latest ChatGPT web app includes mentions of "o4-mini", "o4-mini-high", and "o3".
Reference Link:
https://x.com/OpenAI/status/1910378768172212636
This article is from the WeChat public account "Quantum Bit", authored by Mingmin, published by 36Kr with authorization.
