kooi

I'd call it interpolation on a high dimensional manifold. Convex hull is too simple a shape.

But yes, metaphorically I think that's right.

The other side of peak woke

Yes the problem is very hard. Mainly because high DOF generalization is very difficult.

We have self driving cars because what are the control inputs? Pedal, brake, steering wheel. This already took many many years.

Now for a humanoid robot: An action space that is metaphorically Hilbert. (Physically, yes, obviously)

Also, IMO, LLM's can aid the development of robots, but do little beyond a planning, human control interface. Below that it's the domain of control and the solution will be the correct combination of classical, neural, and real time optimization based control.

All the bad-ass biped robots that actually look natural? It's PID controls wrapped with control barrier functions constraining the QPs that are being solved in real time.

But that's annoying to derive per-application. So we'll need neural methods which can be learned (while being constrained by a priori knowledge of dynamics). My hunch is that the Yann LeCunn type of jepa models will be how tasks can be learned.

Me too

I started manual coding again at least 1 day a week.

Its good for the brain, but man it's really slow.

And the AI results prove it: Over the past couple years I've been able to bring several work projects to completion single handedly.

Our R&D company is now handling more clients and we are returning better outcomes/products faster than before.

So...what to do? Results don't lie.

However I will say LLM's can't write multi threaded double buffers by them selves. Probably because they are so application specific.

But I think the reason I can go do fast is because I went so slow before. I have an intuition I can transfer into an LLM. Question is if that intuition will become sloppy as time goes on...

I used to work there. The product is legit with a well established production line. Hundreds of aircraft in supply awaiting validation and testing.

I wonder how much his stock vested in his 4 years. Could be he vested 10%, is happy, now will move on?

Man marketing really screwed a great engineered product.

4k resolution mouth camera?? That's why we go to a dentist every 6 months because I wouldnt know wtf im looking at.

"Oh there's that canker sore I feel. Huh, at least I can see it"

Bruh, you don't need a camera for gap detection. Just stick an IR sensor, detect change in light. Boom, $0.05.

Poor buddy looks like he's got the shakes

From the video:

Range: 125 miles all electric

       500 miles with hybrid system (electric + conventional turbo prop)

Some high quality electronics projects here.

Its because tk does not have a central event loop built in. You can't use modern programming techniques without building up a significant scaffolding to get multiple UI elements updated based on an events generated in various places. It requires an observer pattern + a self defined callback event generation scheme to collect events (singleton pattern).

Sure you can make this module and there are probably things out there, but you need to wire it up yourself before you even get to the application logic.

Worked on a NASA ground control system written in Tk. It was a great way to learn about classic design pattern because we had to roll our own event reactors, observers, plugins.

Honestly it feels very outdated compared to the reactive type of GUI's we have today.

Maybe there are some nice wrappers now that help...

I'd give a Rust based version a try. Just compile into a binary for your platform and off you go.

Thank you, I think this may get me back on a terminal based development path.

Do you have a specific area you are interested in? That helps narrow down your deep dive hole...because there are so many and you can't do everything immediately.

I think with more agentic code comes the responsibility of more documentation, more architectural understanding.

Perhaps agentic code is the equivalent to a C compiler outputting what previously was the domain of assembly engineers.

Neat. If I'm ever job searching overseas this will be great :)

Termux is great. I have my phone on the same tailnet as my home server. Easy to SSH into it, start codex in terminal, do some work/server adjustments, log out.

Seems a bit over bloated?

We're using Rust with embassy on the esp platform and its quite wonderful.

We have all peripheral modules isolated and have non-peripheral modules compatible with an embassy desktop scheduler and mock peripherals.

Harnesses are good enough now that you can get them in a loop with the actual hardware.

The next step is to hookup a Saleae and have the agent test against actual signals.

Sad and yet so funny at the same time. "I could not read my own logs..." XD I'm so tempted to pitch in 9 bucks.

Let it die ;)

Cool? What's the engine?

We like Sentrux because it's based on score generated from syntax tree parser runs.

How? What are you smelling?

His article on the AI bubble analysis is pretty water tight. https://www.oaktreecapital.com/insights/memo/is-it-a-bubble

Immich + tailscale is great. Setup a little DNS server and now your computer can go to your photos.<lastname>.It's great.

Setting up backups: Ja. Just setup rsync between a couple computers, good to go.

What we see is equivalence to 5.6 Luna. A guess: OpenAI anticipated an open weights release on par with Luna and they dropped price dramatically.

Great writeup, cheers.

Agreed, this like asking a chainsaw to carve a wooden spoon. Impressive it can, but definitely not the right tech to scale.

LLM needs to setup an image classifier to use as a tool call.

skill issue /s

How does this compare to zotero and why would I use it over that?

No wonder they have been gating even reddit post reads. Very lame. On the one hand I get it, they don't want LLM's to scrape their stuff.

On the other hand, come on... that's sure one way to loose traffic.

Because of the increased supply of managing remote devices.

The tradeoff with limited graphics is more expansive platform support.

For example, we're doing some Layer 2/Layer 1 networking processing on an embedded linux environment. We can have a nice TUI displaying average SNR, various waveforms, connection status. It's great.

Also, when you do lots of work with remote devices, having something like termscp is wonderful for file transfer.

Agreed, very cool article.

> I always struggled to understand practical applications.

Me too...and specifically with non-smooth dynamical systems. I always felt some better knowledge or intuition on my part could lead to understanding the practical application.

It's one reason I left academia... wtf this is cool, but wtf is this useful for?

The closest practical application I can fathom in this theoretical realm is optimization. And welp, since that underlies all AI, that's pretty important. Oh and cryptography.