T. R. Shyam Sundar

x by 10 problem solver - past performance * and mentor / :)

[T. R. Shyam Sundar @trssnd]

@thoughtfullless @NeilHudsonexp Thanks. Glad you added this clarification about Hestenes' view which is the context I was missing.

[T. R. Shyam Sundar @trssnd]

@thoughtfullless @NeilHudsonexp How do you say Deutsch describes a separation of philosophy and physics? Asking because I've been thinking he does the exact opposite, interleaving both in his fabric of reality (literally, not the book which I've not read, but I've read his other one, The Beginning of Infinity).

[Ryan Fleury @rfleury]

Critically, CLI vs. GUI is a false dichotomy. CLI is merely one small subset of the set of all GUIs. It is restricted from using certain portions of the space of all GUIs. In order to claim CLIs are superior in all cases, you need to believe that all non-CLI portions of the space of all GUIs are less efficient for humans to use. That is an absurdity.

[KS Sreeram @kssreeram]

The top AI labs talk about “long horizon tasks” that can keep making progress for multiple days before they eventually lose all coherence.

Humans on the other hand have been making continuous progress for millennia!

And we don’t lose coherence in the process.

Not only that. We do the complete opposite.

We _gain_ coherence and keep becoming better over time!

No one knows how to make an artificial system do that. It’s a major open problem. Perhaps harder than even the Rieman conjecture and other great open problems.

[T. R. Shyam Sundar @trssnd]

@kssreeram @Lidinwise @leecronin Wasn't sure about the extent of the analogy about long-horizon tasks as my mental model of LLM and related developments is very poor. This makes it abundantly clear.
https://t.co/lQqQkxz0tQ

[KS Sreeram @kssreeram]

The top AI labs talk about “long horizon tasks” that can keep making progress for multiple days before they eventually lose all coherence.

Humans on the other hand have been making continuous progress for millennia!

And we don’t lose coherence in the process.

Not only that. We do the complete opposite.

We _gain_ coherence and keep becoming better over time!

No one knows how to make an artificial system do that. It’s a major open problem. Perhaps harder than even the Rieman conjecture and other great open problems.

[T. R. Shyam Sundar @trssnd]

@rep_movsd ..immutable state transitions."

I was not remembering the name of the trick. Monad it is. Don't bother unless it's useful for you. You're not dumb. It's just that these are not relevant unless they come knocking on your door.

[T. R. Shyam Sundar @trssnd]

@rep_movsd Again besides the main point, neither pure functions nor practical implementations of Turing machines right?

I think where we are stuck is, I've been thinking they're equivalent all along although I'm also a foreigner to FP tricks for equivalence learned more than a decade back.

[T. R. Shyam Sundar @trssnd]

@rep_movsd "Instead of manually passing the state parameter through dozens of nested functions, the State Monad automatically threads the state through a sequence of computations. You write your logic as if modifying state imperatively, but under the hood, the monad safely manages the..

[T. R. Shyam Sundar @trssnd]

@rep_movsd Again besides the main point, neither pure functions nor practical implementations of Turing machines right?

I think where we are stuck is, I've been thinking they're equivalent all along although I'm also a foreigner to FP tricks for equivalence learned more than a decade back.

[T. R. Shyam Sundar @trssnd]

@rep_movsd Since neither of us are interested in pure theory, skipping exploration of equivalence in FP, especially FP languages.

Got to learn about CompCert though. I'm told it can be used to convert any non-UB C code to Coq. And that converting Coq to Haskell is trivial. *Told by LLM.*

[T. R. Shyam Sundar @trssnd]

@rep_movsd Besides the main point, pure functions a.k.a Lambda Calculus is Turing complete right?

[T. R. Shyam Sundar @trssnd]

I would rather say the animation is what made me understand what I had not earlier as the narration I think assumes an understanding of how ultrasonic scanners work which in hindsight is obvious but wasn't when going through the narration.

[T. R. Shyam Sundar @trssnd]

After going through the entire narration once, the animation from -2:49 for a minute, or even a few seconds actually, beautifully summarised it at a high level helping the understanding stick.

https://t.co/JLDHTkwkmE

[Midjourney @midjourney]

A technical dive inside our new "Midjourney Scanner"

[Daniel Hooper @DanielcHooper]

To get lowest possible mouse latency in a macOS app:
1. use a CAMetalLayer
2. layer.displaySyncEnabled = window_is_fullscreen;
layer.maximumDrawableCount = 2;
layer.presentsWithTransaction = YES;
layer.opaque = YES;
3. mach_wait_until the last possible moment to render so you have the most up-to-date mouse position
4. predict future mouse position based on the last several positions

example code linked in replies

[T. R. Shyam Sundar @trssnd]

Clearer: Still there doing *something else like* that.

[T. R. Shyam Sundar @trssnd]

*Still* there, *doing* that! :)

"The predictions were often aggressive on timing but almost never wrong on direction. And the original direction, written down in 2002 as the company’s mission, was to make humanity multiplanetary. So the board tied his pay to the mission itself."

[Elon Musk @elonmusk]

Make our Sun sentient to understand the Universe and extend the light of consciousness to the stars

[T. R. Shyam Sundar @trssnd]

@DipshikhaGhosh @svembu https://t.co/66WJSgtZl9

[Conjecture Institute @ConjectureInst]

Enemy of Civilization: Scientism

Scientism is the false idea that scientific knowledge trumps all other kinds of knowledge, that science alone can answer all of our questions.

~Conjecture Institute Fellow @arjunkhemani & President @ChipkinLogan

[T. R. Shyam Sundar @trssnd]

More importantly the cloud cover even if it doesn't actually rain.

[T. R. Shyam Sundar @trssnd]

Hah, finally the reason summer surprisingly ends before solstice around here is here. The monsoon rains!

[T. R. Shyam Sundar @trssnd]

@ishan130493 @DavidDeutschOxf > Humans are intelligent because we’ve achieved universality in the domain of explanatory knowledge.

> That means.. there is nothing in this universe that is impossible for us to figure out, unless it is prohibited by the laws of physics.

https://t.co/LhNMqkoVp2

[KS Sreeram @kssreeram]

GPT and Claude are 100% tools.

The problem lies in people misunderstanding intelligence.

One popular definition I’ve seen is:
The ability to perform most economically valuable tasks.

But this falls apart when you look closer…

Compare pre-industrial society to today. The vast majority of economically valuable tasks from the pre-industrial world are now performed by machines.

Does this means the machines from the post-industrial age are intelligent? Most definitely not.

What then is intelligence?

I subscribe to what @DavidDeutschOxf says..

Humans are intelligent because we’ve achieved universality in the domain of explanatory knowledge.

That means.. there is nothing in this universe that is impossible for us to figure out, unless it is prohibited by the laws of physics.

Universality in any domain is necessarily a binary jump. It is or it is not.

Consider computers.. making ever larger and complicated circuits doesn’t result in a microprocessor (which is computationally universal).

In the domain of explanatory knowledge, GPT and Claude are like large complex circuits (non-universal), and humans are like microprocessors (universal).

Making an artificial system universal in the domain of explanatory knowledge is an open problem. No one is making headway on this problem because most don’t understand universality.