Kelsey has a really good lightbulb moment here about platform engineering.

"if you had to do all the deployments for the entire company what questions would you ask of the development team?"

That's your api, your platform, this is your product as a platform engineer. It's not images, docker, terraform, hcl, yaml, kubernetes, It's building out the right api for your company to deploy its products effectively.

https://www.youtube.com/watch?v=HdUbTyvrfKo&t=429s

timestamped

I did not realize all the places to be considered as AI water usage. Hank goes deep highlighting all of the sources he is aware of, most reports leave off a lot of these sources, some reports go maybe too far adding sources that may not make sense depending on the question you are asking.

As someone that runs computers with gpus in their house, and watching LTT make AIO installs on GPUs I've wondered what would AI use water for, now I understand that its a lot. No where near agriculture, but a lot.

Unlike running a gpu in your house, potentially with a closed loop AIO, data centers are filled with hardware making heat and it all must go somewhere. Current technology has this done with evaporative cooling, i.e. its not a closed loop, the water goes into the sky.

He goes on to point out that its not just the data center, using water, but also chip fab and power plants.

Something I hadn't put a lot of thought into is the type of water. While a lot of agriculture and power applications do not use municipal water, a lot of data centers do, putting excess strain on water treatment.

Something I find interesting is that Altman is doing the same thing here that he does on his financial numbers. Stating they use almost nothing on inference. Missing training, missing training for models that turned out to be a flop and never made it anywhere, but they learned how to make better models from it. Missing chip fab, construction, transportation, and power. The way report numbers bases on a small subset of their entire business feels wrong.

Anyways Hank does a great job at saying this is not a simple answer, its complicated. Things are moving so fast that its hard to be accurate, its hard for infrastructure to keep up. There's a lot more to think about than the water use on a single query.

Age verification hitting bluesky?? At least its not yet requiring your govt issued id or anything, but stepping that direction. I don't know how I feel about age checks, does it actually protect kids when parents aren't involved? I can't say anything there, but it really does feel like its about ready to hurt the rest of us, requiring us to whip out ids and personal data for anything done online. This is a real problem that is hard to solve, and reasons why it has not been solved yet.

Deprecation warnings are so easy to miss, ignore, become numb to. Creating tools and processes to catch and address these issues is important. I'm surprised such big projects let deprecations just hang around for years.

damn this is a rough one. A users entire home directory removed by claude code from an rm command.

        
rm -rf tests/ patches/ plan/ ~/

Reading the first half of that command it LGTM. If you had approved rm, you are hosed. If this is inside a larger script its running, you really gotta read close. This one still feels pretty obvious, but I can imagine some bash doing some nasty things I miss if I read it and understand it let alone glance at it.

I'll take this as a reminder that I really need to be paying full-ass attention to agents, and moving towards a better sandbox for them, something in docker, maybe something like distrobox that is a magic wrapper over podman that just gives you the things you need for what it does. Something that starts up with access to start web servers, run agentic cli of choice, see project, git commit. It feels like the right thing has a lot of what distrobox does, but distrobox has too much and would be prone to this using it as I've used it in the past.

This looks like a really good low cost option for some workholding. There is never a shortage of workholding in the shop and everything has a place. Having something low cost that you can have a bunch of makes a lot of sense. Maybe you still need a super scucum unit for really clamping the shit out of something, but this easily covers most use cases in a garage workshop. I want to build it.

Moore's Law is Dead pitches a pretty ingenious sku for the new gabecube aka steam machine. I fully support repairability and ewaste reduction. most of these components have not had MAJOR improvements in years, hence his channel name. There is a possibility here that Valve could ship with their unique hardware, (apu, psu, case, ports, networking) and let you bring your own ssd and ram from an old device that you might not use anymore. I love this idea. At the same time it feels like entering the star wars universe where there are no more new manufacturing and everything is cobbled together from old hardware made long ago.

What a heart breaking video to listen to. I'm trying to do a better job of being positive right now. I'm trying to look at the world in what I have control over (not much more than my attitude about it). AI is killing so much right now I'm trying to look at it as the good tools the engineers made it to be. Ownership is dying around every goddamn corner. Hats off to Edison, this guy gets it. We need more companies like this taking a stand for the average person who wants to make it out there.

What a great campfire story Casey stumbled into. Whether any of this is true few will ever know, but its very reasonable that a race condition and a stalled job to apply configuration caused by someone who left the company 10 years ago caused an outage. I find it hilarious that they call this guy he answers, yup I still know the password, but how do I know you're legit, I'm not just handing out the password. Casey did a stand up job telling this story.

Linus is [[ techbrophobic ]] like the rest of us. This is such an unexpectedly mild take from him. I expected some threat to the mother of the vibe coder, but he gave a pretty great middle of the road take. The industry sucks, it smells off, we know a lot wrong with it, it feels like theres a lot more wrong than we know. But the tools that its making are really good when used in the right ways. They are not a replacement for anything, they are assistive. They can lift someone from not knowing how to code to making a small webapp for their use. Someone who wants to write backend and give them a decent front end, someone who whats to write front end and give them a decent backend.

Great take from someone with more experience than most can ever dream of having, worth a listen.