Hacker News Reader

Comment by pants2 | Hacker News Reader

Much of the value I get from ChatGPT is based on it's memory system that has a good profile on me. For example if I ask it to find me something to eat at X restaurant, it will already know my allergies, dietary preferences, weight goals, medications, other foods I like, etc and suggest an item based on that, all workout me explicitly telling it.

Moving from ChatGPT to Claude I would lose a lot of this valuable history.

View on HN

Or you could type up "My allergies are X, Y, Z, I have the following preferences..." etc. and put it into whatever chat bot you like. Obviously this is a bit of a pain, but it probably constrains significantly how much of a premium ChatGPT can charge. You might not bother if ChatGPT is time and a half more expensive, but what if it's 3x as much as the competition? What if there's a free alternative that's just as good?

I pay for the $200/mo ChatGPT, to me that's insanely cheap compared to the value it provides. Better pricing is highly unlikely to make me switch. If a competitor were able to sustain a lead in model intelligence/capability then I'd consider it.

> Or you could type up "My allergies are X, Y, Z, I have the following preferences..."

Why use a car when you can just walk?

tbh most americans could do with walking more

Even simpler, ask ChatGPT "How would you describe my dietary preferences and restrictions?", maybe throw in some personality guidance of "You are training my executive assistant".

Still not much of a moat, especially with data portability.

In the EU/UK you might not have rights to the memories right now, but you've rights to the inputs that created those memories in the first place.

Wouldn't be too hard to export your chat history into a different AI automatically.

Have you tried asking ChatGPT to output everything it knows about you to a format easily digestible by LLMs? Does the memory stick between model switches?

I suspect it does RAG on my previous convos, so this wouldn't work fully, but to some extent yes that would help.

I like to ask each llm what it knows about me. I feel like I could take that output and feed it into another llm and the new one would be up to speed quickly....

You could literally ask it to write out everything it knows about you into a form usable in a CLAUDE.md file, put that in the directory you're using e.g. claude code in, and boom.

No moat.

Reply on news.ycombinator.com