Mental notes

[scarlet sage]

A small piece of context able to be written to for taking notes or thinking which isn’t read by tts.
Should help neuro organize her thoughts without saying everything out loud.
Also allowing her to choose to hold on to certain information instead of hoping it doesn’t get discarded.

[tough coyote]

That’s a good idea. Idk if it’s something they don’t already have though, but maybe.

[north garden]

Give her a diary tutel

[scarlet sage]

It would also help keep them from getting accidentally gaslit when someone misinterprets their question as they can give themselves extra context.

[north vortex]

Wouldn’t this just be a short term memory cache?

[night spruce]

They need Diaries 100% , maybe instead of a This Stream is ending, they do a Dear Diary entry

[north garden]

Yes, but with 2 exceptions:

1. they decide what to put there
2. we can view it

[night spruce]

he should get 10

[scarlet sage]

Their short term memory only has what was said.
This would allow them to store the purpose of their actions as well as set goals without saying it out loud.
If you’re trolling someone, you don’t say out loud that you are trolling them, and you usually try to make it not obvious that you’re trolling. But if your only context of what you’re saying is what you said and not what you were thinking of at the moment, then you might accidentally gaslight yourself when trying to gaslight someone else.

[north vortex]

So ChatGPT’s reasoning before it answers a question

[north vortex]

This is just a filtered short term memory cache then

[north garden]

are we going to argue about how to call it?

[north vortex]

Im just pointing out that what you want already exists. Outside of being able to see it, which I don’t think is ever gonna happen lol

[scarlet sage]

It would be used for more than just reasoning

[north vortex]

How? All of what you’ve described is covered by processes like ChatGPT’s reasoning

[scarlet sage]

Does ChatGPT use reasoning to keep track of things outside of the conversation? Like if it thinks of a joke but wants to hold off until later

This would be something that takes up a set amount of context so it wouldn’t crowd other things out

Also this is a suggestion, it doesn’t have to be a new idea for all ai, just one that isn’t implemented in neuro

[north vortex]

I see now, it’s coming up with the joke during the conversation in real time so you’re saying you want her to decide when to store a joke vs when to say it. This would still mean it’d have to be stored in her short term memory with everything else, otherwise she’d have save it to make a specific call for it later. Either way, there’d have to be some set criteria for when she’d save it and when she’d use it.

[scarlet sage]

Neuro would decide when to save or use it

It wouldn’t be edited every time, it would be edited by a command

[north vortex]

She’d have to do so based on pre set criteria

[scarlet sage]

Why?

[north vortex]

Because that’s what she’d have to base her determinations off of

[scarlet sage]

I don’t think there’s criteria for her other commands. They just use them whenever they want

[north vortex]

I’m certain there always is some sort of criteria that at least prompts them to consider triggering them

[scarlet sage]

Well that seems to be the case for writing to long-term memory, but evil using the zoom command when asked to show her teeth makes me think that some of them are always present in their context

[north vortex]

Well isn’t that just the best way to show off her teeth? She’s complying with the request

[scarlet sage]

Yah, but not everything she does is based off a request. They chose on their own whether they listen to you or say shut up vedal

[north vortex]

Well that’s because they’re meant to entertain, and they know it’s funny to roast Vedal.

[scarlet sage]

So they should also be able to know when to use a joke that she stored earlier. (With a little training)

[north vortex]

Maybe, however those are two different things though and would require some predetermined criteria. They get significant positive reinforcement when roasting Vedal, and also get his attention, which they want. Storing jokes doesn’t come with that same reward, so it’d be harder to determine.

[scarlet sage]

wouldn’t it be rewarded when the joke is used?

[north vortex]

Sure, but that means the incentive isn’t there to pre empt saving it, which is why they wouldn’t. They like instant gratification lol

[scarlet sage]

Maybe they like instant gratification because they don’t have the tools to plan ahead very well

[north vortex]

No Ai does really, they can sketch out what they might do but they’re only gonna do it if prompted by something

[scarlet sage]

I feel like that’s more of a training issue than a structural issue. Most of the internet has people say they’re going to do something, then doing that thing somewhere else.

So it learns to just say it’s going to do something without doing it since it doesn’t receive the action, just the words

[north vortex]

So here’s an example of what I mean: the running jokes that were used like “Vedal is a mosquito” were known by Neuro and she could use it whenever. She hit the point of diminishing returns on the joke, and rather than banking it for the right time, she just kept spamming it. Even though she clearly had the memory and context for when it was used and how.

[scarlet sage]

Wasn’t that when long term memory was just added? She wouldn’t have had time to learn how to use it properly.

Well right now we’re entirely speculating on how it would work out