donquichotte

Reminds me of my favorite abstract, from John Doyle's paper "Guaranteed Margins for LQG Regulators".

https://improbable.com/wp-content/uploads/2011/10/ThereAreNo...

"But Wolfgang, could these Regulierungen be umgangen by hosting TiefSuch Blitz on our very own Rechneresysteme?"

I also love the gaslighting of some models, like ChatGPT mixing in words with cyrillic letters and when asked about it answers: "it can look as Slavic to the eye" and "sorry that it came across as Russian"

Feels a bit more light hearted and humourous, more like Brazil?

I bought one of the 60W Chinese Laser cutters, 40cm x 40cm, with RUIDA controls for the steppers. It's a decent machine and it runs with Lightburn. I get shocked a little bit from the HV supply when touching fluid in the cooling loop, and there is no interlock preventing the laser from firing with an open hood, but it cuts well and the mirrors are adjustable. It was about 700$ for the machine and 700$ for shipping to Europe in 2023.

Just keep in mind this is not a desktop machine to run in a flat. You need some infrastructure like a cooling loop and air exhaust, so this thing best lives in a basement or hobby room.

Claude Code stopped working in tmux for me at some point. Which is of course more of a CC problem than a tmux one, but maybe OP was trying to work around such issues.

Damn 15 AK47 bullets per token

I was not trying to make that kind of argument, but rather a geopolitical one.

Retaining world-class talent and home-growing industries is no small feat. In Europe, we have been struggling with this, and I am not even entirely sure why.

My conspiracy theory is that this is the new space race, and the CCP encourages this to show the world what Chinese engineers are capable of, and tank the Anthropic/OpenAI valuation bubble as a desirable side effect.

There is really no good reason to ever use I2C, except if that's the only interface your peripheral supports. Debugging a stuck I2C bus, and attempting to un-brick the bus for a stuck slave (disabling ISRs, unhooking DMA, reconfiguring pins to GPIO, clocking the slave out of its coma and then doing everything in reverse) is just not worth it.

EDIT: I was hoping for HaLow WiFi!

> I can think of the tedious task of finding certain functionality within a codebase. I usually can't beat an AI agent harness at this task today.

Yup, I remember "racing" the AIs to figure things out in codebases just a year ago. Today, I have no chance. Whether it is due to degraded reasoning capabilities on my part or better models, I don't know.

I also got an email from Anthropic: "We're updating our Privacy Policy". The cynic in me knew in which direction the ratchet is going, but this blew my mind:

> As part of our measures to keep our services safe and secure we may ask you to verify your age or identity, and we've described what we collect and how.

Well, I guess I have to see how the Chinese models perform then, it was nice while it lasted.

Same here, I expected an "openclaw powered google glass", leading to a "Black Mirror"-esque world full of glassholes regurgitating LLM output at each other.

Beautiful illustrations!

Interesting, do you have any examples of the latter two categories? Looking for a replacement for my Cherry-Keyboard.

Are you using an external provider for image generation or running something locally?

In Europe, this is mandated by law.

https://en.m.wikipedia.org/wiki/ECall

I recently installed NetBSD on a Thinkpad T430s, it is surprisingly snappy and usable, except when firefox is running.

With i3wm, tmux and helix with clangd it is a capable development machine with a surprisingly modern feel, only the compile times give away that the machine is from the early 2010s. Even WiFi is working, and pkgsrc is fast and well-stocked with the newest software.

cnlohr is in his own league, this guy is bordering on genius.

Dissatisfied with the state of the vendor SDK for CH32V003 microcontrollers (ultra-cheap RISC-V MCUs), he created his own [1], which is a pleasure to use. He also has a header-only RISC-V emulator that runs Linux (and Doom!) [2], and hacked an ESP32 to emit valid LORA frames with clever use of aliasing [3].

[1] https://github.com/cnlohr/ch32v003fun

[2] https://github.com/cnlohr/mini-rv32ima

[3] https://www.youtube.com/watch?v=eIdHBDSQHyw

It's an interesting question, but only looking at energy expenditure probably does not cut it.

I hike a lot and started running races in alpine territory (e.g. this one [1]) about a decade ago and think if you take into account fatigue (mental, muscle, tendons) and exposure to the elements (sun, rain, wind, snow), a light running pace can definitely make you arrive at your destination less exhausted than walking the same distance, given that you can move confidently in the given terrain.

So it is a multi-dimensional optimization problem (as opposed to only optimizing for energy expenditure) and very dependent on how comfortably you can move at the given velocities.

[1] https://www.youtube.com/watch?v=o-12ghcODMM

I agree, but I have seen an implementation of a middle ground (Scaled Agile for Enterprises, SAFE) that combined the worst of both worlds, with quarterly, one-week, all-hands-on-deck mammoth PI (product increment) planning sessions down to the story point (whatever unit that was, apparently it has magical self-calibrating properties) that were thrown out of the window 4 weeks into the PI.

> There was no incentive to carefully consider where to place a sticker. Players didn’t have enough agency to advance the plot through their individual action. This limited creativity to simple patterns like tiling similar stickers or forming lines.

Ah, the game turned into a big corp job!

STM64 subsystem for STM32

If I understand it correctly, ~20MWh is for the full size model at 700m depth and with a radius of 15m (prototype radius is 1.5m and it is located at a depth of 100m).

And building concrete spheres that can withstand the pressure of a 700m water column is probably an interesting design challenge on its own.

My thoughts exactly. It makes the engineer in me go mad if I see projects like Energy Vault [1] getting massive funding that could be used to try and develop technologies that make sense. Thankfully there are some people who see through the charade [2].

If you are into this thing and looking for an even more stupid idea to store energy, I present to you the StEnSEA [3]. Rolls right off the tongue, right? It is a hollow concrete sphere that is lowered to the bottom of the lake. Pumps then remove water from it, creating a vacuum. Letting the water back in and using the pumps as generators, the energy is reclaimed. Curiously absent from all documentation of this project is the amount of energy stored. I did some back of the envelope calculations a while back and it is 3.8kWh, for a multi-million-euro prototype!

[1] https://www.energyvault.com/ev1

[2] https://www.youtube.com/watch?v=iGGOjD_OtAM

[3] https://www.iee.fraunhofer.de/de/projekte/suche/2013/stensea...

For anyone interested in the structure of the software that ran Ingenuity (and some hardware design aspects, such as the use of commercial off-the-shelve parts), there is an awesome and critically underwatched video of Timothy Canham explaining everything:

https://www.youtube.com/watch?v=mQu9m4MG5Gc&t=7s

There was also Wurmz!, a networked multiplayer version of Liero. Some history and binaries of various version as well as the source code can be found on this website [1].

[1] https://mental-reverb.com/wurmz.php

So the inverse of the final scene in Patrick Süskind's "Das Parfum" is plausible?

1) collect women's tears from donations (kickstarter.com)

2) spread them as a fine mist at G20

3) world peace

Edit: maybe this is what Hugh Laurie was singing about here: https://www.youtube.com/watch?v=Q8chs2ncYIw

> I see no reason to see why you should expect it to work in harder, nonlinear settings

I'm not so sure about this, maybe this is where the ML approach could outperform (in terms of estimation accuracy, not compute time) the traditional EKF and UKF approaches, by learning the nonlinear system dynamics?

This sounds very hand-wavy, and it is, because of my lack of understanding. For me it is just not immediately clear that if an optimal algorithm for the linear case cannot be matched or outperformed, that is also necessarily the case for nonlinear dynamics.

EDIT: And as mentioned above, the KF is optimal if certain conditions hold, e.g. additive, zero-mean, Gaussian noise on state dynamics and observation. In reality, you may have a multiplicative component of the noise nor non-zero mean or fancy noise distributions, and it would be interesting to see if these can be learned.

Nice talk! Travis Goodspeed, who she mentions helped her decapping the chip, has some very interesting reverse engineering talks, e.g. a tear-down of a commercial glucose monitor:

https://filestore.fortinet.com/fortiguard/research/glucose-b...