coop

Imagine if we only got 1/3 of the channels in the advertised TV package.

"Sorry, we only sell up to 3000 channels, but can't guarantee you'll see them all"

I still think it's crazy that ISPs can sell me "3 Gigabit internet" when I usually don't even get 1.

They should have to advertise minimums.

Yes, that is exactly what they think, and it's exactly as stupid as we all know.

You push a commit. You wait. A runner picks it up. You watch logs scroll. Something fails. The error looks like someone fed a stack trace through a paper shredder and then set the shredder on fire. You add a run: env step to see what’s going on. You push again. You wait again. A twenty-minute feedback loop for a one-line change. You do this fourteen times. This is your afternoon now. You had plans. You were going to go outside. The afternoon belongs to the CI now. It has always belonged to the CI. You are only now perceiving this truth.

www.iankduncan.com/engineering/...

Slack does it too, and I get a lot more of those than emails.

Have you found that masquerading as an AI lady has made you more successful? Asking for a friend?

I think it's a somewhat fundamental issue with today's models. They are trained on Reddit and Tumblr as well as GitHub and Stack Overflow, and I don't think they understand the difference.

But if you give up Reddit as a source you lose a lot of your models worldliness.

I just realized it applies to repositories and libraries too. I'm not instinctively looking for signs that a human was involved at all. A project being older than 2 years old is a good sign.

Tell me again why we want this future?

Interesting to see the charts align with my observations. It's presumably made way worse by all the hidden tokens we aren't shown from web results and system prompting.

I just learned that it's called "context rot".

Crappy AI blog posts have permeated all my interests; gardening, cooking, woodworking, programming, etc.
When I go looking for information, I have zero faith in anything I find.

It's pretty upsetting actually 😞

And I suppose future AI are going to be trained on this slop. Entrenching it forever.

I'd hope I could blame someone else too.

Moderation is probably the hardest part of building any user content platform.

If it's done right it's invisible, too much and you hate free speech, too little and you hate children. Impossible to make everyone happy.

Bsky is nice for now, but it will go through it all soon enough.

Four or five buckets worth. All thanks to this amazing wood chip maker.

Wood chips next to the wood chip maker

Apparently, I've been insisting for a year that we not buy shelves for our house. Supposedly I said I was going to build them.

Apparently, that's too long to be stacking crap on the floor.

Anyways, I didn't get any coding done today, but I sure made a lot of wood chips.

Has anybody figured out how to work with LLMs to problem solve? They all constantly forget earlier important context.

Me: Hey, how do I do X with Docker?

LLM: Just do A!

Me: That'll be a problem because of Z!

LLM: Okay, just try B!

Me: That wont work either because of Z!

LLM: Okay, lets do A!

Your own server

For better off or worse I set up #Postgres inside my Docker Swarm cluster tonight. Including a sidecar to backup to S3.

I'm getting pretty close to having a fully deployed stack. Hoping to launch for real by the end of the month!

What a tease! I guess I'll just keep using em for now.

Docker Swarm secrets being immutable has been a real pain for me.

Would you recommend an external secrets manager?

Basically, it was a new concept, and I didn't feel like figuring it out. I'm a big ol' nerd too, so I can imagine most laypeople would give up at that point too.

When I decided AI wanted to move off Twitter I looked at Mastodon and Bluesky. Mastodon just had a slightly too high barrier to entry, though I admire its goals.

Ha! Fair points. Honestly... I might do it anyway. I don't have a very intense use case case, and having everything as little Lego blocks is very appealing.

LLM: This will work I promise... Okay, this time it will work I am certain... This time for sure...

Me: This seems wrong?

LLM: You're absolutely correct, thank you for pointing that out. This time it will work...

Tis a good feeling

Thanks, I looked hard at it a while ago, it looks fantastic and almost made me move to k8s just for it. But I after dabbling I decided it was more than I wanted to manage.

What's wrong with running PG in production with Docker? I know it needs several config tweaks, but it seemed okay otherwise.

Always remember to containerize your nuclear arsenal

I did manage to get the #RaspberryPi connected, something to do with DHCP. But I can't get in by SSH yet. Will try again tomorrow, can't wait to get #Tailscale running, I wonder if I can get the boot script to set it up.

I should mention that I don't have a monitor or keyboard, so the iteration cycle involves reflashing a painfully slow sd card, messing with the boot partition from my laptop, and then restarting the pi.

3AM, and I'm still trying to get this damned first gen #RaspberryPi to connect to my home network. It must be angry about being left in a drawer for a decade.