“Small,” non-generative, BERT-like models, would probably be more appropriate.

Anthropic does sentiment analysis with regex in Claude Code though ¯\(ツ)/¯

I see an em-dash on a comment in MainActivity.kt on line 278, so I’m guessing it was used extensively. Also, a “→” character on 291.

Impossible. Signal said they have no choice but to use AWS for this kind of thing.

And offshoring to vibe coders. Saw an article the other day about Indian Amazon devs (well, probably the managers) saying that being forced to use Amazon’s in-house models instead of Claude was hampering their productivity.

I think I’d install proxmox on all machines because there is a proxmox provider for Terraform. Then, manually create the VMs, and to learn the barebones, use kubeadm to set everything up, and kubectl to manage it. Once comfortable and knowledgeable with that, start messing around with Terraform and Ansible.

Just don’t use AI coding tools then? Studies show they make people less productive anyways.

It’s been a while since I messed with home automation, but ESPHome was amazing to program ESP microcontrollers (i.e. you most likely wouldn’t have to write any code). You can use ESPHome devices with both Home Assistant and Openhab (using MQTT, IIRC). The last I checked, it was easier to program your own functionality in OpenHAB than Home Assistant.

For servers, I usually choose the distro with a version with the EOL scheduled furthest into the future. Usually that means Ubuntu (Server) LTS.

I have not personally worked on large projects using functional languages. I know they are popular in finance/trading.

Been a while since I’ve used Scala, but I remember Scala being much more focused on functional programming than Kotlin.

Postgres is basically an open source version of Oracle DB. Much more featureful than MySQL. I believe Oracle bought MySQL just to kill it.

All the open hardware and software and ecosystem was pretty cool. It was cool you could just buy hats, or whatever they’re called, to add functionality, rather than designing a custom PCB and spending hundreds of dollars to get a few boards made and populated. I’m not a fan of their software stack or their choice of uC’s, but they did make it easy to just kind of plug stuff together in hardware and software.

Port forward/poke holes in firewall + dynamic DNS.

I’ve been going down this rabbit hole myself. Already set up a solar Meshtastic node and MeshCore repeater. Kinda cool, very low bandwidth and pretty unreliable though.

It’s my understanding that encryption is illegal on amateur radio bands. I’m thinking about getting a license anyways; looks fun.

HaLow, BATMAN, Reticulum and stuff like that also look cool, but I haven’t messed around with those yet.

I think radio will always have bandwidth/congestion problems. It’s like everyone within range is using the same “wire.”

I also like overlay networks like Tor and I2P, but it’s possible those will eventually be blocked or made illegal in many countries, if governments keep heading in the direction they seem to be heading.

It’s ok. It seems a bit ad-hoc to me compared to some of the higher-level academic languages I like (Haskell, Scala, Clojure, Chapel, etc). It’s much simpler than C++, and more expressive than C though.

You can boot Pis off an USB HDD or SSD. I think there are specific hats for that as well. But yeah, backups, at least of configs, are nice.

It is much easier to maintain software with static type checking.

Debatable

I always just considered AI as a CS field that included things like planning, pathfinding, logic and symbolic reasoning, and ML, of course. Most of it is built on optimization and constraint satisfaction algorithms.

You can create communities on an existing instance just by clicking “Create Community” in the top menu; no need to host anything yourself. At least you can on the instance I use; not sure about lemmy.world.

Kinda reminds me of this: I built the most expensive CPU ever! (Every instruction is a prompt)