Lovah Spaghets!
@spaghets.bsky.social
"In these “latent spaces” concepts become directions, conceptual categories become clusters, and reasoning unfolds through mutually informed transformations of sequences of high-dimensional vector patterns."
Those directions in that space are knowledge about the world.
Those directions in that space are knowledge about the world.
September 23, 2026 at 6:01 AM
"In these “latent spaces” concepts become directions, conceptual categories become clusters, and reasoning unfolds through mutually informed transformations of sequences of high-dimensional vector patterns."
Those directions in that space are knowledge about the world.
Those directions in that space are knowledge about the world.
LLMs process representations in high-dimensional vector spaces where meaning is encoded in geometry.
aiprospects.substack.com/p/llms-and-b...
aiprospects.substack.com/p/llms-and-b...
LLMs and Beyond: All Roads Lead to Latent Space
Essential concepts for understanding AI today and prospects for tomorrow. (Bonus footnotes: the maths of exponential orthogonality & updated citations 30 May 2025)
aiprospects.substack.com
September 23, 2026 at 6:00 AM
LLMs process representations in high-dimensional vector spaces where meaning is encoded in geometry.
aiprospects.substack.com/p/llms-and-b...
aiprospects.substack.com/p/llms-and-b...
It's literally how LLMs work.
September 23, 2026 at 5:57 AM
It's literally how LLMs work.
Busing doesn't help this. These kids are way behind even when they're in the exact same classrooms as kids who are doing well.
September 22, 2026 at 5:48 PM
Busing doesn't help this. These kids are way behind even when they're in the exact same classrooms as kids who are doing well.
It's your argument. 🤷
Is producing unreliable output an argument for or against AI behaving like humans?
Is producing unreliable output an argument for or against AI behaving like humans?
September 22, 2026 at 5:44 PM
It's your argument. 🤷
Is producing unreliable output an argument for or against AI behaving like humans?
Is producing unreliable output an argument for or against AI behaving like humans?
And I think if we could roll back time a few minutes, your brain would do the exact same things it did the first time. I'm not even sure what argument you're making.
September 22, 2026 at 5:41 PM
And I think if we could roll back time a few minutes, your brain would do the exact same things it did the first time. I'm not even sure what argument you're making.
>They work exactly according to their algorithm.
So what? Why do you believe an algorithm is incapable of thought?
Yes, autocomplete can think if it's backed by a sufficient process. You're just restating the Chinese room argument.
So what? Why do you believe an algorithm is incapable of thought?
Yes, autocomplete can think if it's backed by a sufficient process. You're just restating the Chinese room argument.
September 22, 2026 at 5:38 PM
>They work exactly according to their algorithm.
So what? Why do you believe an algorithm is incapable of thought?
Yes, autocomplete can think if it's backed by a sufficient process. You're just restating the Chinese room argument.
So what? Why do you believe an algorithm is incapable of thought?
Yes, autocomplete can think if it's backed by a sufficient process. You're just restating the Chinese room argument.
You say that like it matters at all. How does knowing what each node is doing exclude their collective state being able to infer facts about the world?
September 22, 2026 at 5:34 PM
You say that like it matters at all. How does knowing what each node is doing exclude their collective state being able to infer facts about the world?
Evidence of what, exactly? These models are unquestionably intelligent, are capable of reasoning about problems that humans can't solve, create art and music. This is reality today, and anyone who uses them knows it.
And I got my CS degree in 1996.
And I got my CS degree in 1996.
September 21, 2026 at 9:55 PM
Evidence of what, exactly? These models are unquestionably intelligent, are capable of reasoning about problems that humans can't solve, create art and music. This is reality today, and anyone who uses them knows it.
And I got my CS degree in 1996.
And I got my CS degree in 1996.
Reposted by Lovah Spaghets!
One thing I'll say: a whoooole lot of people are staking out strong philosophical views on AI with a false sense of security, serenely unaware that the capabilities that would severely undermine their confident claims already exist.
September 21, 2026 at 7:29 PM
One thing I'll say: a whoooole lot of people are staking out strong philosophical views on AI with a false sense of security, serenely unaware that the capabilities that would severely undermine their confident claims already exist.
The fact that you're admitting that humans have the exact same failing as the ai - hallucinations - as an argument against the AI is very interesting.
September 21, 2026 at 9:41 PM
The fact that you're admitting that humans have the exact same failing as the ai - hallucinations - as an argument against the AI is very interesting.
Except it the exact same argument, just used in a way you don't like.
September 21, 2026 at 9:40 PM
Except it the exact same argument, just used in a way you don't like.
But not via reasoning and induction/deduction.
September 21, 2026 at 9:38 PM
But not via reasoning and induction/deduction.
So what? It still solved the problem when no human was able to and not just by trying every possible combination or something equally boring.
September 21, 2026 at 9:36 PM
So what? It still solved the problem when no human was able to and not just by trying every possible combination or something equally boring.
When the prompt is "solve this unsolved math problem" and it does it, that's amazing and not just repeating the question back.
September 21, 2026 at 9:34 PM
When the prompt is "solve this unsolved math problem" and it does it, that's amazing and not just repeating the question back.
This is not brute forcing like they do when trying to crack large primes. This is more like hiring 1000 mathematicians to work on the problem at the same time.
September 21, 2026 at 9:31 PM
This is not brute forcing like they do when trying to crack large primes. This is more like hiring 1000 mathematicians to work on the problem at the same time.
You're using an opinion piece to argue the science?
February 25, 2026 at 7:24 PM
You're using an opinion piece to argue the science?
Ai art, and AI in general, doesn't need any defense. Almost everyone is using it every day for work, and that's not going to stop.
Chatgpt is the 5th most used site on the Internet and rising.
Chatgpt is the 5th most used site on the Internet and rising.
May 27, 2025 at 4:57 PM
Ai art, and AI in general, doesn't need any defense. Almost everyone is using it every day for work, and that's not going to stop.
Chatgpt is the 5th most used site on the Internet and rising.
Chatgpt is the 5th most used site on the Internet and rising.
The AI industry won't be killed by this, it will simply become the Chinese AI that everyone uses.
May 27, 2025 at 4:56 PM
The AI industry won't be killed by this, it will simply become the Chinese AI that everyone uses.
This is how you lose your user base to a competitor who uses AI.
May 27, 2025 at 4:55 PM
This is how you lose your user base to a competitor who uses AI.
Buying a Tesla has never been more completely out of the question.
March 17, 2025 at 5:29 AM
Buying a Tesla has never been more completely out of the question.
If it's a fake issue, then why are the democrats in the Senate so out of touch with the members of the Democratic party on the issue?
March 7, 2025 at 11:31 PM
If it's a fake issue, then why are the democrats in the Senate so out of touch with the members of the Democratic party on the issue?