Memory is not needed

Over the past month I have tried both OpenClaw and Hermes, which I installed so my brain would start coming up with use cases. For now I have settled on the latter, paired with Discord, and I am having fun with it. I still use ChatGPT in the web app and mobile apps mostly as a Google replacement and a simple question-answering machine.

In both places, almost at the same time, I started getting seriously annoyed by how memory works out of the box. You ask it a very specific, self-contained question, and it starts with: "As an Android Tech Lead, this might be interesting to you because..." or "You asked about X a month ago, so Y is probably not what you need...". That is not what I asked you about. Why are you wasting my tokens?

In short, automatic memory in context is terrible. Its job is only to create the illusion of intelligence for normies, but in practice everything useful is extremely fragile, and everything useless is also harmful.

I disabled memory completely in ChatGPT settings. In Hermes, I switched to a setup where I periodically review what it has accumulated and approve only the things that are genuinely permanent.

It still seems like the only correct path if you want to keep controlling the system. If only you knew how much verbose slop I found in there about myself, on topics that interested me only in the moment or were tied to very specific tasks.