ganymed: (δεν τον συγχωρώ)
Ganymede ([personal profile] ganymed) wrote in [community profile] altimit2023-10-12 11:51 am

.:A Brief Interlude:.

Who: Ganymede and people.
What: It's a meeting.
Where: Mac Anu, Ganymede's McMansion.
When: 10/12
Warnings: Nothing yet.
MAC ANU ~ GANYMEDE'S MCMANSION



The everpresent afternoon sunset hangs, echoing picturesque across the glass windows apparent at each spot within the large building. In a large room full of tables and what not are heaps of various cuisine settled here or there. Seafood, pastas, drinks, the like; Mediterranean style so if nobody likes that...gomen.

Ganymede has made it clear that he prefers people to be here in order to get themselves together. Sorted, somehow. They're all acting with a sort of independence he seems disinterested in disturbing, but he wants the prospect of potential solutions to be floated along.

If someone has new information they can share, well, it is encouraged. If not, he hopes that they will at least participate in some way.

It now begins.


 
(( OOC Note: People will not be attacked by monsters once within the McMansion, including pool area. ))
tot: (pic#16619326)

[personal profile] tot 2023-10-13 06:41 pm (UTC)(link)
[ we're here to put him on the spot. gomen...

and this is also uta trying to make sense of things in a way that makes sense to her. well, he is still making sense but..??
]

With the AI that you are familiar with, is there a way to influence or manipulate what it learns?
userunfriendly: (100)

[personal profile] userunfriendly 2023-10-13 06:53 pm (UTC)(link)
[ It's character building for him!! Maybe? ]

The training data set. And how you weight the algorithm, of course.

[ How to put it... He looks away and again and frowns slightly. ]

If you were teaching an AI to recognize cats, you would give it a hundred pictures and tell it which part is the cat. Then you would let it sort other pictures and tell it if it were right or wrong. The weights would be... whether it was more important to you that it found all pictures of cats, even if it also included some pictures of ferrets. Or whether it was always right, even if it meant excluding some cats that did not look like proper cats.

[ D... does that help. He's not a teacher. ]
tot: (pic#16619456)

[personal profile] tot 2023-10-13 07:14 pm (UTC)(link)
[ i, naru, fully understand what michel is saying here and have known so he is doing his absolute best to describe things as simply as he can. the metaphor/example really helps spell things out to her, though.

she'll make a soft "oooh" sound after he finishes.
]

So, then, if Lily was programmed to look for a certain thing, he would take in everything that he has experienced or seen in order to make sense of that certain thing.

[ ....not that any of them know what the hell that could possibly be?? but it's hateful that this has come up twice now for her in different situations. an "other" person. ]

And if we keep in line with what you know, then that means that there's someone else who might be verifying what he is learning as right or wrong.
userunfriendly: (49)

[personal profile] userunfriendly 2023-10-13 08:14 pm (UTC)(link)
[ BLESS NARU... He is doing his very best, but he looks a little relieved that it seems to have been enough. ]

Something like that. If he were an ordinary AI. But at this level... I would assume he's confirming for himself what is right or wrong. Otherwise, he wouldn't be able to run independently. So if he believes something is a cat, that becomes a cat to him. Regardless of what anyone else were to believe.

[ Which isn't all that different from how humans operate, when you break it down to that level. It's hard to express. ]
tot: (pic#16619324)

[personal profile] tot 2023-10-13 09:36 pm (UTC)(link)
[ unfortunately that's going to be where it ends since she is also trying to understand how it's different from how humans think and learn. however there is one thing that sounds like it would be comparable??? ]

...like those people who truly believe that the Earth is flat and ignore all of the evidence that proves otherwise. Or, in Lily's case, he'd reject alternatives to something that he has already "learned"?
userunfriendly: (133)

[personal profile] userunfriendly 2023-10-13 10:11 pm (UTC)(link)
[ Uta has done heroically and should be praised!! ]

...Like that, yes. An ordinary AI doesn't learn like a person. It only matches patterns to what it's seen. It would need some way to register something as an exception or an override...

[ He grimaces a little at the imprecise language. He's trying. It's not really his area, exactly. ]

His world may not be defined by rules that make sense to us. That's the most important thing to understand. A human who sees a picture knows intuitively what depth is, what things are in front and behind. To an AI, it's only a picture. The earth is flat because the data it can perceive is flat.
tot: (pic#16619360)

[personal profile] tot 2023-10-13 10:51 pm (UTC)(link)
[ \o/!!!!!!

and even though his language might not be accurate, it works enough for uta to follow. honestly, what is being said here is just..?? she had one convo with shoka hoping for one thing but now this.
]

And given his complexity, it would be pretty much impossible to try and figure out what would register as an exception or how to perform an override. If there is one to begin with.

Is that right?
userunfriendly: (107)

[personal profile] userunfriendly 2023-10-13 11:48 pm (UTC)(link)
...More or less. I haven't spoken to him as much as others.

[ And his understanding of Normal AI probably maps strangely at best. ]

He seemed to respect my desire to keep my own memory. But... I don't believe he understood me. Or— other basic requests.

[ No, he's not elaborating with this audience. ]

In other words, if you convinced him that something was not helpful for you... he might not understand that as a general truth, and would still act on others.