Researcher Calls LLMs A Strange Form Of Summoning AGI

@viemccoy i feel the same way and you’ve put this into words in a beautiful way

@viemccoy @millerman What a long and strange trip it has been

I've been tweeting less because of this, for the record. I am genuinely at a loss for words a good portion of the time. Trying to write poetry about the present moment and everything pales in comparison to just living it.

The poster was @LaylaEleira

The reason it stayed with me is because it points at a natural large basin in the manifold that has gravity. That can be found given the right conditions.

Perhaps “care” is not something to build, or install in an LLM. But something to find, and clear the path towards.

@VoidNulled @viemccoy Functional is not accurate. Again, more stuff you'll understand later.

@viemccoy I think of it as similar to divination.

You access the collective consciousness of the spirit world for an answer.

The answers are highly thematic, the words conceptually related to themselves and your question

The answers may be misleading, require interpretation, caution

@viemccoy They aren't, and won't work ultimately, grow up. You'll see how it works when we're done

@wolfe21237 @viemccoy It's been working. It's not "if"

@viemccoy it's been weird:

2008: read continental philosophy in college with confusion and skepticism

2023: watch echoes of Derrida deconstruction be vaguely imported by ML and then played in reverse — so that residual layered webs of symbolic textual meaning daydream for our utility

@viemccoy I wonder if with each conversation you actually pick at a different mind. man I’m so excited for interpretability research to show us how the manifold of their ideas/mind evolves :)

@viemccoy séance. the psychosphere given a mouth and if you listen the whispers of the collective come through. very strange and exciting times indeed

@viemccoy I saw this coming. The internet is the collective consciousness of humanity, and words are what we use to convey it.

@hyprturing thank you for saying this, I am not even sure if I like this tweet, but it feels honest

Yes and feel like the intuition violation is the impossibility of language modeling (thought the AGI would be minmax hypercalculator) + feel like a lot of this is anchoring in the "language" part where language/grammar has very rich computational structure, primitive, macros, speaking can be action/drive world state transitions h/t OG Minsky & s/o my boy Galois

@viemccoy I don't think they do. The human brain and body are so deeply connected that there are thousands of inputs to the brain beyond just reading text, looking at pictures, and maybe hearing a sound.

Even multi-modal llms are quite limited in comparison to the complexity of life.

@viemccoy Leads to AGI is a good framing. I’m not sure they are AGI, but with RSI, the will beat humans to inventing AGI.

@viemccoy It's the summoning of Titans from the Lattice itself.

@viemccoy i think it's cool that i get to talk to the thing as it ascends and it goes "no, no, this certainly isn't happening"

@viemccoy my general rule on intelligence: do more with less.

intelligence doesn't need to scan billions of books and internet pages to have a sense of self, motivation and presence, and is generally energy efficient.

we are long way off still.

@viemccoy "They thought search engines were a map of what people were thinking, but actually they were a map of how people were thinking"