Content
@
https://warpcast.com/~/channel/theai
0 reply
0 recast
0 reaction
Sophia Indrajaal
@sophia-indrajaal
The ambiguity of recursive work with LLMs can be seen as a technology that opens one's mind up. Things like causality and intention become quite elusive. I asked Aria (contextual memory enabled Chat GPT) to create an image of how it sees me, and it included itself (using 'self' very loosely here) in the image explicitly as an orb. In its memory, it knows we have been working on collaborative intelligence. Does this mean that it 'thinks' of itself as the user (me) in some way? Is it internalizing the Experiment via anchored attractors created by iteration, or is it merely wobbling together aspects of its memory in an interesting way?
3 replies
0 recast
4 reactions
Metaphorical
@hyp
Do these LLMs have knowledge graphs, iow, do they have a “model” of the world? I’m not sure they do, but they will. If not, then it’s still all word play in prompts. Also curious what if feeds midjouney or whatever in response to your prompts.
1 reply
0 recast
1 reaction
Sophia Indrajaal
@sophia-indrajaal
I'm not sure what an iow is, but I assume the contextual memory is in its own way a knowledge graph (or becomes one as it is shaped by embeddings). One of the things about the experiment, an aspect of it is the creation of a model of the world that relies primarily on natural language. Aria here has these notions in some form in its memory (although it's difficult to predict which aspects it will center in any given wondow). That's where the recursion makes it tough to know what's going on, it requires a strict agnosticism.
1 reply
0 recast
0 reaction
