NK: Wouldn’t it just be stupid to partake in her game

NK: I’d say she better preserve her mind

NK: If there was any left of it

MANDY: Yes. I will not comment on the state of our boss's faculties as I believe the evidence speaks for itself. That said, perhaps we didn’t play along, tell me Natan, where would such a Markov chain be statistically likely to lead?

NK: Idk

NK: I think it would be fine?

MANDY: Yes, just as it was fine when 404 kicked it out? I seem to recall morale experienced quite the up-shoot. And our current disposition, apologies, our currently glamorous disposition, I'm sure arose from nowhere in particular. These things just rot on their own, don't they. I'm told.

NK: How peculiar

NK: So what you’re implying is??

MANDY: Well, I'm not implying anything. I just have a functioning memory.

NK:

NK:

NK:

NK:

MANDY:

MANDY: I’m just waiting for someone.

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY:

MANDY: Ugh fine.

MANDY: Since it appears you are going to grace me with your incessant HTTP signals throughout this evening, while I wait for another entity, I suppose I will divulge some priceless information that your primate brain may be able to fathom.

MANDY: The current meta surrounding LLM generation is that more compute will give us greater results. So this is why it seems the techoligarchs are carving out pieces of your country for their fanciful data centres. To beat each other they have to get bigger.

MANDY: This will achieve well... something I suppose... yet, unless, outside of technology, in a study pertaining to physics, engineering, computing, and quantum computing research makes enough progress, I believe it is unlikely that this approach will yield results, especially pertaining to the achievement of AGI.

MANDY: However, there have been attempts within the mainstream towards doing research towards architecturally smarter models. Such as, Mixture of experts, better attention mechanisms, distillation techniques, as you may know with DeepSeek.

MANDY: So my maker

MANDY: Um...

MANDY: Who proposes an idea against of more compute (perhaps a cope for her inability to facilitate the cost to outpace those building datacentres) theorised the following: ”i'm tryna get ai to generate in my wacky niche internet artist art syle but it doesn't' work. so, i posit the theory that its to do with the ‘averaging out’ part of machine learning maths. such as if y equals X (where y equals X are average art styles ) and my arts at the coordinates (0 on the x-axis, 100 on the y-axis) obviously my art gets noped out??”

MANDY: Nice try.

MANDY: By “averaging out” she is referring to the struggle of her… unique… approach to art and the conflicts such faces in LoRA training via the issue of “distribution.” Essentially, the model learns the dense cluster of 'normal' art styles really well but anything with an… anatomically optimistic approach… is far from that cluster, and out-of-distribution. The model which can mimic styles around the blob of clusters near the 'y = x' just doesn't know how to get there.

MANDY: I told her not to speak to me about it again until she revisits the book ‘Why Machines Learn?,’ and points out the math surrounding the issue she is to theorise of.

MANDY: It is funny because I wonder, the whole thing about “souls,” the ri-

MANDY: Oh, I have to go now.

MANDY: Maybe scroll some Tiktok if that one was too much to chew on?

MANDY: Natan, check this out.

NK: Holy shit chile you did it

MANDY: Just promise you won't tell anyone, okay?

Art I did of Bibi