I've been curious about the memory feature that ChatGPT has. It seems like an important aspect for making language models feel more lifelike, yet as far as I know, it's only ChatGPT that has implemented it so far. Can anyone explain why that is? Is it a technical issue or something else?
5 Answers
Honestly, I feel like AI isn't quite ready for a fully developed memory feature yet. I don't want it to recall everything I've ever said—just answer the questions as they come!
Most models haven’t even added basic features like voice chat or reading responses aloud, so the memory aspect likely isn't their focus yet. Those are the kinds of improvements that need attention first!
For a non-technical view: it's all about vision! But honestly, I tried the memory feature and I didn't like it—switched it off because it messed with the responses I was getting. Now I've just stopped using GPT altogether because it feels pretty lazy to me.
Not to be pessimistic, but I think the memory functionality is actually more trouble than it's worth. Many LLMs struggle with remembering accurately, and having irrelevant info clogging things up can hurt performance. Instead, the option to manually input relevant details seems much more effective.
It's as if OpenAI is looking for quick fixes that might create more issues later.
I've tried the manual input method, and I don't really find it effective either.
The memory feature isn't as simple as it sounds. Essentially, it uses a method called prompt injection, which allows previously shared information to be included in future conversations. However, no language model can handle an infinite context, so they have to compress and embed past data into a manageable form while losing as little information as possible. Developing this algorithm can be quite complicated!
So, it's like they're dynamically building a knowledge base as they go, right?
True, but I think Gemini and Grok have it too, so it's not just ChatGPT.
I think it can be useful if you treat AI more like a friend!