🏅 Top traders

@Bayesian , @Gabrielle , and I all discussed this market at great length and we decided unanimously to resolve it No. We took into account every comment and piece of evidence available on the market.

This seems like it should resolve NO to me? ChatGPT can be given additional information to remember, but it doesn't just automatically have the context of "all the information in all of my interactions with it".

@Gabrielle I see the argument from @Jacy below -- is that feature still around? It seems like it would be close to counting at least. But I mostly agree with your thought here.

This feels like a case where we may need to do the 3 moderator panel thing. If any other @traders have context to share, do so soon.

@Eliza @Bayesian maybe knows more? Have you ever used this system?

@Eliza I probably don't know more but

so a model released by OpenAI can recall all the information in all of my interactions with it?

had definitely not been achieved before 2025. they pretty much had the tech to do that (RAG) but not worth it, the memory feature was very minor.

I see the argument from @Jacy below -- is that feature still around?

That feature is still around, may have changed with the latest change to 4o in the last few days (not informed ab that, but doesn't matter to this market bc that's a 2025 change), but yeah that still existed on jan 1st and i don't think would come anywhere close to counting? If anyone disagrees please explain why

@Bayesian Much appreciated. It looks like we have 3 mods here so if we all decide the same way we can resolve.

@Eliza @Gabrielle @Bayesian Memory is still around and allows for recall "across all our chat history." I would agree that constantly available RAG would be more intensive recall, and putting all the history into the context window (perhaps with some compressed attention algorithm) would be even more intensive. When the market was made, I think the memory feature would be the most common interpretation of the market title/criteria, even though that ended up feeling somewhat basic by the end of 2024.

@Jacy that's a good point

@Bayesian @Gabrielle I found a webpage describing this feature:

https://openai.com/index/memory-and-new-controls-for-chatgpt/

I think the length and style of the content is within the normal expected resolution effort for moderators in this case, let's make sure we all read it and understand it before we decide.

If any traders have suggestions of OTHER resources related to other features not mentioned yet, share those also.

I’ve read through the article on memory, and used the memory feature several times before. Without knowing what FUH meant at the time of creating the market, we only can go off of the comments and description. The description clearly states “Will this change by 2025 so a model released by OpenAI can recall all the information in all of my interactions with it?” It doesn’t say “information that it chose to store before” or “some information from some past interactions”, but “all information in all of my interactions”. I think this is setting a clear line that the memory feature is not trying to surpass, so I vote to resolve NO.

@Gabrielle For context, an example feature that would make me resolve YES would be a feature that allows ChatGPT to search in past interactions or to load past interactions that it thinks are relevant. But with memory, not only does it have to choose at the time to store it in the memory, but there’s a maximum length for the memory that can be stored.

@Gabrielle

I read the linked page, comments here, etc. I think I understand the feature sufficiently to give an opinion.

"Across all my interactions" -- it can recall some information for use in all future interactions, but that's not "all" the past information. Only a tiny selection.

"all the information" -- does not meet this bar

I would resolve No.

@mods This can resolve YES, thank you!

@10thOfficial Can you explain why?

I think this qualifies, @firstuserhere.

@Jacy I'm thinking about it:

... model released by OpenAI can recall ...

@firstuserhere They roll out with custom GPTs because it doesn't actually work and they wanna see others fail before they implement it themselves

@firstuserhere just to be clear, this is the main ChatGPT interface, not a third-party "GPT" from their app store. I think this qualifies. However, the feature seemed to be taken down shortly after I made that comment, alongside pop-ups saying their servers were under high demand and ChatGPT being unresponsive at times. So I realize you can't test it yourself.

Must this be through the openai interface?

@theshortbread yes, or the api

@firstuserhere through the API you can already just give it all prior conversations in the context length?

@theshortbread oh yeah, obviously. Through the interface, yes.