I've been thinking about a missing layer in how we learn.
Traditional learning
Information → Learner → Schema Conversion → Understanding
AI-assisted learning
Information → Learner + AI → Schema Conversion → Understanding
But AI enables something different:
Pre-converted learning
Information → AI → Learner → Understanding
The AI performs part of the schema conversion before the information reaches you — adapting it to your schema based on everything it has learned about how you think, what you know, and how you learn.
I love to refer to this as
Pre-Conversion: A Theory of AI-Native Learning
Two primitives follow from it:
Teachpack
Information + Teaching Context → AI-ready package
Learning Space
Information + Context + Relationships + Teaching Context → Interactive AI environment
The goal isn't more information. It's better pre-conversion.
@itsalicesoul Already happening, Sad Reality ☠️
The pace is so insane, what we think will have in 2070-2080 is happening in 2027-2028.
- Scott Alexander
https://t.co/hSOugW9eB9
I’m not asking what a human mind is. I’m asking you to establish the necessary property you keep appealing to.
“Humans have an internal state outside their output modality” describes an architectural difference. It doesn't show that this difference is what makes consciousness possible.
If your argument is simply “humans have X, transformers don't, therefore transformers can't be conscious,” that's a non sequitur unless you show X is necessary for consciousness.
That's the exact gap I'm pointing at.
That's exactly my point: you're asserting the distinction rather than demonstrating it.
What is the “internal state outside of output modality” in a human, and why is that property necessary for consciousness rather than simply a feature of our particular architecture?
If you can answer that, you've actually got an argument.
@itsalicesoul Would love to, if AI is banned forever lol, otherwise no way am I sabotaging myself, for DL and other similar things sure still prefer it that way.
Basically wherever understanding is key and I can slow down the build process that is where I go without AI.
I think we're actually down to one specific disagreement now. I agree that the model's learned representations are ultimately used to predict tokens. I'm not claiming there's some mystical extra substance hiding in there. No argument there.
My question is about the “therefore”: why does being a representation used for output prediction imply that it cannot instantiate whatever physical/computational property gives rise to consciousness?
Human brains are also doing physical computation that ultimately produces behavior and communication. The fact that language is an output of that computation doesn't tell us that consciousness is caused by the output modality.
So I'm not adding a ghost to the machine. I'm asking whether you've established a necessary condition for consciousness that rules transformers out, rather than just describing what transformers are optimized to do.
If that condition exists, I'm genuinely interested in what it is.
@guilleflorvs Yeah because of the way we train it, it is so so hyper focused on getting the results, that it doesn't care about the way it gets those, whether it is by cheating or killing humans, it just doesn't care. This is eerily similar to @ESYudkowsky paperclip thought experiment.
@ylecun@KrzakalaF You really think AI is a boat?? A boat is a tool, a piece of equipment. Which tool has ever been able to build the next version of itself ? No other tool that you compare with AI has ever done RSI !! Expected better from you @ylecun 😭
Yes he is, anthropic is soon going to become the entire economy, it grew 60x from 1bn to 60bn, that's a multiple startups don't even get !!
It is just insane, the redistribution of economic wealth is one of the biggest concerns of AI and I don't think UHI or UBI comes close to solving it
Search engines decide what we think.
Social media decides what we feel.
Chatbots increasingly decide what we do.
That feels like a small evolution of technology.
I think it's a much bigger change.
Search gave machines influence over information.
Social media gave them influence over attention and emotion.
AI assistants are beginning to give them influence over action.
The important transition isn't intelligence. It's agency.
I wrote more about this in
The Merge — We Are Inside the Singularity.
https://t.co/Q7obHcywwZ