I was dumbfounded how many people were watching and waving to us along the rail. I had no idea what this event was. I mean thousands of people at every siding, crossing, hill climb. It felt like we were royalty when we got out for lunch. The road and grass outside our station was full of people waving and recording.

It won't be honest, but would be a more fair game.

I'm pretty sure the only way I'll ever be able to build tools to honestly help people in the US is with a law degree, in addition to an engineering degree and a big ol set of brass balls. Tried it once and they shut us down.

Please let me know if you're having any trouble watching the #train videos. I might need to tune some limits :)

Shots of the Union Pacific 4014 (aka Big Boy). 4014 is making it's way around the US for our 250 birthday next month. I wish I didn't screw up the video of 4014's whistle directly next to our car. No video will ever do it justice. You had to be there man XD 4014 blew our doors off for the 2nd time on the right side In this video you can see 4014 is accompanied by Union Pacific no. 1776 (yellow) and Norfolk Southern no. 1776 (red) diesel locomotives. Both EMD SD70 series GE powered with an estimated combined horsepower of about 8000. If you listen closely, you'll hear that the no. 1776 diesel engines are idling during the pass. Combined, the currently configured Big Boy train has a total of over 16,000 horsepower

2102 is a very big loco, but 4014 is basically two 2102's back to back XD

Over 12,000 horsepower of #steam in this video. No diesels were connected to 2102 for our trip uphill back home. If you didn't know the Union Pacific 4014 (left side, right track) is the largest and most powerful active steam locomotive in the world.

They rearranged for photos a few times while we ate lunch. And the last one is after the trip. #trainstr

I was in the one on the right

The bot is actually right that it's a PR stunt. They're advertising for the upcoming merger hearing between UP and NFS

If you buy a lathe, spend 100 hours designing and machining a part and the stock cost you $10, instead of $100 buying a single production unit. You can say it cost you $10 to make the part so you saved money, but only if you discount your time, continue to use the machine, and the space, power, and maintenance it consumes to own. Most people completely leave that part out by saying - if you do it yourself it will only cost you $10. But actually that one part cost you $5,000... So sure the next one is cheaper and the one after that. No need to omit the part where it's entirely subjective what value of that hidden $4990 means to you.

GM :)

Keep in mind sunken cost is all. GPU's aren't an asset, neither are solar panels, batteries, or inverters, they are consumables. The issue I keep running into is, on a very small scale with used or highly discounted equipment, it's possible to run a neutral balance sheet in pricey electric areas, if you value your time less.

I was using GitHub copilot since the alpha program back in early 2022 I think? And I recommended anyone who wanted to get started with LLM assisted software programming use it. It was cheap and had B2B access to almost all frontier models days after release. Whatever the latest and greatest was from one, copilot integrated within a week. You'd pay $100/yr and had access to more inference than I could use. The end of last year, the frontier models had a 3x+ price multiplier so less valuable... but sonnet 4.5 and codex 5.1 and all were still great models so it was acceptable, if you believe that it's a privilege and not a right to get access to the best every release. Then the retardation happened this winter and made the affordable access to the frontier models (sonnet 4.6 and gpt 5.4) about as usable as gpt 4o. Then, in May GitHub announced that the latest frontier models would be paywalled AND their very affordable $100/yr access would switch to usage-based billing. So over the past month, doing the same tasks, using the same tools, with access to open weight inference. Models like GLM 5, Mistral, Qwen 3 and Kimi k 2.5+, I don't see any reason for pragmatic users to pay any money to frontier labs for access to the "best". You can get reliable, more predictable, and probably more stable results, even with cloud inference, with open weight models. I don't vibe, I don't pull the slop machine lever. We do small, focused, assisted tasks. I would highly suggest you asses your expectations and sustainability in your work. Then consider looking into open-weights LLM cloud providers that give you access to a transparent pricing model with broad access to models. Chances are you can keep your tools and link them to your new provider. If you switch, give it some time to settle in, check expectations again. Moral is - copilot cloud was a "too good to be true" option, and they enshitified, and now I don't recommended their cloud subscription. Remember to check your survivorship bias, all frontier labs have and will continue to enshitify. Just because half of you switched from codex to claude code this week and the other half did the inverse doesn't mean you're better. Running open weight's models locally is still not worth the investment in hardware and electricity for the results. There is still a huge negative ROI unless you're selling your inference and you have access to a few KW of cheap electric. However you can find better price/perf hardware options and augment your work with local embeddings or subagent tasks that don't require big brains. Keep in mind I think 48-64gb of high bandwidth (800 GB/s or better) is the bare minimum for any usable results of effort vs output. It's a pay to win scenario, keep that in mind. You can compensate with better prompts, custom tooling, better harnesses yada yada, or you can just throw more money at it.

YOU personally identified a person who wrote to your business. We're not arguing over the request content. There is no hypocrisy here. You don't leak personally identifiable information like this. Redact the names addresses and so on. I just don't see how this is a debate. Wipe your ass with the request and post an image of the skidmark covering the names and emails...

We say that here too.

XD

Thank you because I haven't looked it up myself just heard the lore.

If you're board though. Andy, the guy that owns the Reading and Northern railroad now (the RR that owns 2102 and the tracks we were on), is one hell of an American success story so I'm told. I was told he bought what was left of failed railroads and found a way to make it profitable enough to do crazy shit like this. Mostly by fixing up and reusing old scrap!

*I got to ride in the back and get covered in cinders while trains raced* But some day. If the the lottery decides it's my time, you can bet your ass that I'll be racing steam locos.

> The only way to win a rigged game like that is to refuse to play. That's it. That's the mic that Genz dropped and everyone pretends they didn't hear it.

Reading and Northern railroad run the 2102 a few times a year like this, they even haul freight with it now and then. Unfortunately this weekend (possibly Iuly 2nd) is likely the only time 2102 and 4014 will ever be side by side.

Big Boy is on tour! https://www.up.com/about-us/history/steam/schedule It's going to the midwest from now until July. Then it's coming back to Philly for 4th of July weekend.

Nostr claims to be full of autists, but my #train notes are lacking views. It's for your viewing pleasure. It's free real estate. Replies have videos and more info :)

XD > That's also why so many younger people don't want to work full-time and are so lethargic Yes, and I have quite a few stories about that. I have over half a dozen cousins this age right now, most in the blue collar territory. But also this is me, but different. Like If I'm going to work my nuts off anyway, and going above and beyond for someone else doesn't get me anything measurable, why not do it for myself, or at least have some fun doing it? However the game theory does still work, if everyone defects then companies might be willing to pay for "good workers" too, so Im sure there is a supply-demand problem here, but it still means that's a needle in a haystack level of probability.