stevejb

I have a Dell R740xd that is the core of my home lab, with 384 GB of DDR4 ECC. That was by far the most expensive part of the server. I get a lot of use out of it, but it certainly contributes to a higher-than-average power bill. I'm working on solar at some point next year.

I feel the same way. I don't like or dislike Stallman, and certainly disagree with him on a lot of stuff, but I can appreciate what he stands for and for GNU in general.

I distinctly remember my dad telling my "my .emacs file is older than you are." I'm 41!

I don't know enough about Github or Microsoft to say anything intelligent, but I agree with your sentiment in general. And, one thing is for sure - I have used Github for years, pushing many thousands of commits on personal, private projects. I've burned through as many free action minutes as I can, and I use my own action runners as well. I've done this all for free, so on balance I am grateful for them.

I have a Mac Studio M2 Max as my main driver. It's fantastic. I will get the best M5 Studio I can afford when the higher RAM options come out because I am very involved with local inference. I find that with Tailscale I can either use screen sharing if I've got a good enough internet connection or just SSH and Mosh if it's a little bit laggy. That has been good enough for me. A Mac Studio and a Neo as a companion is a really good combination. My laptop is an Apple M4 Air but it basically does nothing except for running ScreenShare and a terminal.

You can run GLM 5.2 at a reasonable quant on a cluster of DGX Sparks. On the link below, the guy is getting decent performance and links to a repo. An 8x Spark cluster can run it even better. A Spark cluster is about as good as it gets for (a) runnable at home, (b) "affordable" hardware, (c) not going to kill your power bill.

https://www.youtube.com/watch?v=nbHOBvLlypY https://www.youtube.com/watch?v=PV89U-PNUNA

Their bar graph showed that in almost every category except for accessories, the weights were pretty much identical.

Yeah it seems like a cheap CNC drill which can drill holes in steel sheets would be something that would last basically forever. Something like this:

TTC3018 CNC Router Machine 3-Axis Engraving Machine Metal Engraver GRBL Control with Built-in WiFi, 775 Spindle Motor, Engraver Desktop CNC for Wood, Acrylic, Plastic, Aluminum, PCB Milling & Carving https://www.amazon.com/TTC3018-Machine-Engraving-Engraver-Al...

Hey I just did a bit of a deep dive into this. Wow, thanks for pointing this out. It seems like 'PEEK' would last the longest, but most printers can't do this.

"Standard estimates for indoor PETG longevity are 10–20+ years under normal conditions, with some sources giving a more conservative 5–10 years. A century is 5–10x beyond those estimates under passive conditions." Wow! Thanks for pointing this out. This is fascinating.

You can print QR code with different heights representing white or black. That piece of plastic would last a long time, and you could encode.....something. You could pint a version 40 (177x177) grid with a Prusa XL easily, and that would last for a few thousand years and be waterproof and fairly resilient to a lot of things. You could probably print a 177x177 grid on a regular Prusa MK4 and just have a color change layer. E.g. layer 1 white, and then layer 2 black. This would be 2,953 bytes.

You could even get a binder, and then print the QR codes sheets as 'pages'. You could print these thin enough to do 1mm sheets. That means ~ 120 of these in a 5 inch binder. So, a 3 ring binder of 1mm sheets of 177x177 QR codes would hold about 346.2 KiB. You could store encryption keys in this way. You can probably push the density well past 177x177 even on a Mk4s pretty safely. I may try this later today.

Isn't this like saying that if better woodworking tools come out, and you like woodworking, that woodworking somehow 'isn't your craft'. They said that their craft is about making things.

There are woodworkers on YouTube who use CNC, some who use the best Festool stuff but nothing that moves on its own, and some who only use handtools. Where is the line at which woodworking is not their craft?

Hey! I built a Lego technic car once 20 years ago. I am fully confident that I can build an actual road worthy electric vehicle. It's just a couple of edge cases and a bit bigger right? /s

Ah interesting. I have been using OpenCode more and more and I prefer it to Claude Code. I use OpenCode with Sonnet and/or Opus (among other models) with Bedrock, but paying metered rates for Opus is a way to go bankrupt fast!

Is this still the case? Is Anthropic still not allowing access to OpenCode?

Most AI agents have a 'bash mode' and, you can use Warp terminal which is terminal first, but easy to activate the AI from the terminal. For example, if you mangle a jq command, it will use AI to suggest the right way to do it.

stephenbarr.ai

Yeah, whenever there are human subjects there is an IRB which is necessary. But, beyond that, we didn't participate in the market in any way. We wanted to see if there was bias there, and how much of it. I think I may have used the word 'value' in a bad way in my description. Not 'value' as in 'can we exploit people?' but value as in statistical significance. E.g. if you applied for a loan and your profile picture contained yourself with a child, did that help you, hurt you, or was it neutral?

Using the Propser.com data set (a peer-to-peer lending market), I used MTurk to analyze the images of people applying for a loan. This was used in a finance research project with 3 University of Washington professors of Finance.

The idea was that the Prosper data set contained all of the information that a lending officer would have, but they also had user-submitted pictures. We wanted to see if there was value in the information conveyed in the pictures. For example, if they had a puppy or a child in the picture, did this increase the probability that the loan would get funded? That sort of thing. It was a very fun project!

Paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1343275

On the show Full House S05E25 the character Michelle was learning to cook. She made tuna flavored ice cream. Uncle Jesse says something to the effect of "It's great that you like tuna and great that you like ice cream. You don't have to combine them."

I try to remember that episode when building tech products. We all like solar. We all like trains. It doesn't mean that we need to have solar panels between the train tracks.

I think that although it could be cool, it seems like train right-of-ways is a particularly harsh environment for solar panels. There's dust, harsh vibrations, heavy cast iron components, and other things right next to a sensitive bit of electronics. It seems like it would be more economical to have a solar farm managed by the train company. This way the panels can be easily cleaned, angled properly, and maintained not in the proximity of giant rolling metal boxes.

> The sprinkler defaults to off to conserve water as the system is potentially unmonitored and a burst pipe could cause issues.

I had a friend in Australia who ran cattle on his farm. Failing open would waste water, but failing closed would mean dead cattle (and hundreds of thousands in losses). It depends on the application.

It isn't clear that 'conserve water' is a reasonable default position. Perhaps 'keep doing what it was programmed to be doing' would be a better position?

Amazon's revenue in 2024 was about the size of Belgium's GDP. Higher than Sweden or Ireland. It makes a profit similar to Norway, without drilling for offshore oil or maintaining a navy. I think they've got plenty of juice left.

I'm still on Carmen's Headline Viewer. I see no reason to swtich.

> "Blue text, while also a widely recognizable clickable-text indicator, is crass and distracting. Luckily, it is also rendered unnecessary by the use of underlining."

Can someone explain this to me? I don't get the 'crass' part of it. I find that having different colors for things really helps me parse things quickly. By removing the color dimension from the equation, it seems to make it more difficult to know what is what.

Isn't that completely circular then? E.g. "if there are no technologies which change $X, then we predict that $X will stay the same."

I use the Warp terminal and I can ask it to run —-help and it figures it out

I just saw Weird Al in concert, and one of my favorite songs of his is "Everything You Know is Wrong." This is the AWS version of that song! Nice work Corey!

> It's more a combination of a lack of a meaningful train system

Doesn't the USA have the world's largest and most cost-effective rail freight network? This seems meaningful.

> Until we get superhuman autonomous coding agents the human in the loop grokking the generated code is still the limiting factor.

They never get tired, work for pennies, can search the internet and your code base, follow rules, and iterate on test cases. This is better than I can do, so by my reference point as a human, the coding agents are superhuman already.

This is super cool. When I was doing a lot of math-related typesetting I found that I got really fast with Emacs + AuCTeX. I have no idea of the state of that project, but if you were fast you go at nearly the speed at which you could write.