Jack Horne
@jackeroni.bsky.social
Data scientist extraordinaire, Lover of dogs and coffee. Still trying to figure out things in life. Check out my newsletter jackhorne.net
Someone purchased si.ai and it’s currently on the market for $1.1B
Buy si.ai | Spaceship
Own si.ai today. Secure checkout and guided transfer support. No hidden fees.
si.ai
October 8, 2026 at 8:36 PM
Someone purchased si.ai and it’s currently on the market for $1.1B
Change in net popularity of <something> X-axis is days before election? Data are indexed to levels 100-days prior to election and forecasted for current year.
October 7, 2026 at 1:31 AM
Change in net popularity of <something> X-axis is days before election? Data are indexed to levels 100-days prior to election and forecasted for current year.
Nice thread and post. But I hate these space filling charts. This one in particular makes it appear like live ticket sales decreased from 2000-2015 while a deeper look shows they continued to grow.
Consider a line chart with a separate line for total in situations like this
Consider a line chart with a separate line for total in situations like this
September 28, 2026 at 2:24 AM
Nice thread and post. But I hate these space filling charts. This one in particular makes it appear like live ticket sales decreased from 2000-2015 while a deeper look shows they continued to grow.
Consider a line chart with a separate line for total in situations like this
Consider a line chart with a separate line for total in situations like this
Shoot, a guy could have a pretty good time in Vegas with that.
September 25, 2026 at 12:54 AM
Shoot, a guy could have a pretty good time in Vegas with that.
Just so folks are clear on the meaning of a proof … a proof would be a generalizable set of circumstances where the N/S equations do not apply, accompanied by a new set of equations.
No one is claiming that that’s what the OpenAI computers has produced.
No one is claiming that that’s what the OpenAI computers has produced.
September 22, 2026 at 7:37 PM
Just so folks are clear on the meaning of a proof … a proof would be a generalizable set of circumstances where the N/S equations do not apply, accompanied by a new set of equations.
No one is claiming that that’s what the OpenAI computers has produced.
No one is claiming that that’s what the OpenAI computers has produced.
A new proof to the N/S problem has not been established yet, or at least hasn’t been published.
what OpenAI did was to brute force computing power to find one instance where the N/S equations faltered. In no sense is that “reasoning.“
www.science.org/content/arti...
what OpenAI did was to brute force computing power to find one instance where the N/S equations faltered. In no sense is that “reasoning.“
www.science.org/content/arti...
How an AI math breakthrough ignited a controversy
OpenAI spent millions to solve the Navier-Stokes problem, but who should get credit?
www.science.org
September 22, 2026 at 6:38 PM
A new proof to the N/S problem has not been established yet, or at least hasn’t been published.
what OpenAI did was to brute force computing power to find one instance where the N/S equations faltered. In no sense is that “reasoning.“
www.science.org/content/arti...
what OpenAI did was to brute force computing power to find one instance where the N/S equations faltered. In no sense is that “reasoning.“
www.science.org/content/arti...
And there must be some activation function that causes the shift at 0 on the x-axis 😅
September 22, 2026 at 2:04 PM
And there must be some activation function that causes the shift at 0 on the x-axis 😅
Forget it Jake, it’s Bluesky
September 22, 2026 at 2:18 AM
Forget it Jake, it’s Bluesky
I don’t have that many fingers and toes.
September 22, 2026 at 12:49 AM
I don’t have that many fingers and toes.
It goes to your original statement on th thread. Linear and nonlinear methods are not fundamentally different from each other. It’s just far more efficient to use nonlinear methods for computation and programming
September 22, 2026 at 12:47 AM
It goes to your original statement on th thread. Linear and nonlinear methods are not fundamentally different from each other. It’s just far more efficient to use nonlinear methods for computation and programming
Yes. An optimization is not a closed system either and thus does not yield finite results. A linear system will still get you to the same approximation. It’s just terribly inefficient.
September 22, 2026 at 12:26 AM
Yes. An optimization is not a closed system either and thus does not yield finite results. A linear system will still get you to the same approximation. It’s just terribly inefficient.
We’re saying the same thing, and I agree with you that it’s the nonlinear aspects that make the models useful.
But, at the heart of the matter, a log transform or even a La Grange optimization can be broken down into a series of linear equations. It’s just less efficient to program it that way
But, at the heart of the matter, a log transform or even a La Grange optimization can be broken down into a series of linear equations. It’s just less efficient to program it that way
September 22, 2026 at 12:15 AM
We’re saying the same thing, and I agree with you that it’s the nonlinear aspects that make the models useful.
But, at the heart of the matter, a log transform or even a La Grange optimization can be broken down into a series of linear equations. It’s just less efficient to program it that way
But, at the heart of the matter, a log transform or even a La Grange optimization can be broken down into a series of linear equations. It’s just less efficient to program it that way
Neural nets do do this though. They just apply the linear transformations on a layered or a nodal basis as you suggest.
In a way you’re both right. LLMs involve more than just linear algebra, but the nonlinear algebra that they use derives from linear algebra.
In a way you’re both right. LLMs involve more than just linear algebra, but the nonlinear algebra that they use derives from linear algebra.
September 22, 2026 at 12:00 AM
Neural nets do do this though. They just apply the linear transformations on a layered or a nodal basis as you suggest.
In a way you’re both right. LLMs involve more than just linear algebra, but the nonlinear algebra that they use derives from linear algebra.
In a way you’re both right. LLMs involve more than just linear algebra, but the nonlinear algebra that they use derives from linear algebra.
You’d probably need pencil and paper 😊
September 21, 2026 at 11:17 PM
You’d probably need pencil and paper 😊
LLMs are not a product of those things (except for the last one) in that these things didn’t produce LLMs. LLMs are just math as Kathryn has been explaining. The first three items are necessary precursors to the use of modern LLMs though.
You’re making my point about nuclear weapons for me.
You’re making my point about nuclear weapons for me.
September 21, 2026 at 10:49 PM
LLMs are not a product of those things (except for the last one) in that these things didn’t produce LLMs. LLMs are just math as Kathryn has been explaining. The first three items are necessary precursors to the use of modern LLMs though.
You’re making my point about nuclear weapons for me.
You’re making my point about nuclear weapons for me.
Oh my, now we’re getting into my lane. That’s math *and* a bunch of highly combustible stuff.
If you want to apply the same to LLMs, you’d have to say that the massive amount of energy to run the machines is part of the math itself. It’s not, but the energy use in an of itself is a little scary.
If you want to apply the same to LLMs, you’d have to say that the massive amount of energy to run the machines is part of the math itself. It’s not, but the energy use in an of itself is a little scary.
September 21, 2026 at 10:39 PM
Oh my, now we’re getting into my lane. That’s math *and* a bunch of highly combustible stuff.
If you want to apply the same to LLMs, you’d have to say that the massive amount of energy to run the machines is part of the math itself. It’s not, but the energy use in an of itself is a little scary.
If you want to apply the same to LLMs, you’d have to say that the massive amount of energy to run the machines is part of the math itself. It’s not, but the energy use in an of itself is a little scary.
Agreed (see my second sentence). Horizontal stacking *could* cause one to conclude a fractional seat prediction.
It’s a reasonable device so that the shape of the chart isn’t too tall and narrow. A less confusing device might be to make each box (or fraction of a box) represent 10 simultations.
It’s a reasonable device so that the shape of the chart isn’t too tall and narrow. A less confusing device might be to make each box (or fraction of a box) represent 10 simultations.
September 16, 2026 at 6:27 PM
Agreed (see my second sentence). Horizontal stacking *could* cause one to conclude a fractional seat prediction.
It’s a reasonable device so that the shape of the chart isn’t too tall and narrow. A less confusing device might be to make each box (or fraction of a box) represent 10 simultations.
It’s a reasonable device so that the shape of the chart isn’t too tall and narrow. A less confusing device might be to make each box (or fraction of a box) represent 10 simultations.
Can you explain how a simulation results in a fraction of a seat? Or are you just stacking columns horizontally to save vertical space on your chart?
September 16, 2026 at 5:29 PM
Can you explain how a simulation results in a fraction of a seat? Or are you just stacking columns horizontally to save vertical space on your chart?
Bob’s Red Mill ‘00 Flour for pizza dough. Ingredients: Wheat Flour. That’s the whole list.
September 6, 2026 at 1:10 AM
Bob’s Red Mill ‘00 Flour for pizza dough. Ingredients: Wheat Flour. That’s the whole list.
They’re just adding more training data and making the model “smarter,” you see. 🤔
September 1, 2026 at 8:34 PM
They’re just adding more training data and making the model “smarter,” you see. 🤔
2008-9 was the middle of the subprime mortgage crisis. There were far more originations and as a result far more defaults. Today many boomers own their houses outright and many others can’t afford to buy, ergo fewer defaults. Trend line is concerning though.
fred.stlouisfed.org/graph/fredgr...
fred.stlouisfed.org/graph/fredgr...
fred.stlouisfed.org
August 31, 2026 at 3:53 PM
2008-9 was the middle of the subprime mortgage crisis. There were far more originations and as a result far more defaults. Today many boomers own their houses outright and many others can’t afford to buy, ergo fewer defaults. Trend line is concerning though.
fred.stlouisfed.org/graph/fredgr...
fred.stlouisfed.org/graph/fredgr...
“Closed course. Do not attempt.”
August 18, 2026 at 3:14 AM
“Closed course. Do not attempt.”
*Sankey, of course, in the first sentence 🙃
August 18, 2026 at 1:40 AM
*Sankey, of course, in the first sentence 🙃