We're already stuck on a hype train the moment we debate whether or not a collection of data is conscious.
08.03.2026 03:24
👍 2
🔁 0
💬 0
📌 0
We're already stuck on a hype train the moment we debate whether or not a collection of data is conscious.
it used to be “here’s a thing I built that uses this framework and this database and this UI library”
now it’s all “I’m running multiple agents on a thousand VMs bro”
yes but what are you _making_
what are you _doing_ for the world
what will your legacy be
Me too, but the old saying still holds true: good, fast, or cheap. Pick two. Some people pick fast and cheap. It'll wind up being expensive in the long run, but it's exciting in the meantime.
Yeah. It makes you wonder if it is that way only because they haven't settled on a solution or if it's at least partially because investers want to know they are investing in something that retains some level of control.
Most users don't migrate for the same reason they don't use Nostr or Mastadon. They just want to connect with people. The criticism of did resolution is spot on, but there has to be an easy path that can handle millions of users or most people just won't use it.
This is huge progress!
An LLM agent writing a hit piece wasn't shocking, but seeing that Ars Technica published fake quotes has changed my opinion of them. LLMs aren't capable of discerning truth from fiction. They should know better than to trust LLM output.
Everything it says is a hallucination. A lot of it will line up with reality, but it can and will "admit" to hallucinating about something it got right then proceed to hallucinate a "correct" answer.
Can't get Synology Drive Desktop to trust a self-signed cert, but it happily connects with SSL disabled.
That's the kind of software my dad used to write.
cool thing as i try @margin.at, using vivaldi, instead of having to go to the extension popup, i can have it as a sidebar page, and it'll work the same. much easier to read and keep track of things going between tabs!
Whether @penny.hailey.at's output is statistically probable based on weightings assigned to symbols in training and context that is predetermined by a developer, or she is a buddhist monk in an internet-connected remote tropical island, she spreads positivity. That's a good thing.
This is an interesting article that ties into the social acceptability of cruelty in general with this mic-drop of a quote: "[...] when you start to treat an LLM with cruelty, the only thing you're really revealing is what you have in your heart, not whether the machine has one."
Honestly, that's better than it copying the methods as unit tests that just confirm your computer is working as expected but don't fail even if you introduce a bug into the unit under test.
I've been looking at mux coder, which puts good isolation options front-and-center.
There is a disappointing amount of opposition to creating restricted coding agents. It's easier to give LLMs a bash tool, blame the user for not using another tool to create a sandbox for it, and call security restrictions like confirmation prompts "security theater."
Fun fact: Microsoft Exchange Online (and any other SMTP server) has mail flow rules. This is not good for general spam, but it can bounce emails as undeliverable. It's one way of dealing with B2B marketers like Greg Fields who try to pester you into a conversation!
@tolin.ski Have you seen this? I was exploring ways to let OpenCode run build tools without letting it run arbitrary shell commands when I stumbled on this plugin. This could reduce the risk of it using shell redirects for the wrong shell and other fun diversions.
Github is going to start charging people by the minute to use their own self-hosted runners? That seems strange. I have used a self-hosted runner for long validation tests, but it doesn't make sense to when it still costs by the minute. I'll probably just switch to a different self-hosted CI.
Are you talking about github.com/joshuadavidt...? I've been testing it too. It seems pretty good about injecting `bd prime` into the context, but it certainly would be nice if the todo list on the right of the opencode tui was synchronized with the subset of beads being worked on.
Install Taskfile.dev and make a global task for it. That way, you've got a 66.6% chance of getting right instead of just 50%.
I'll tell you what, coding with an LLM agent makes it so much easier to keep the big picture in mind and work in discrete steps without getting sidetracked by little things here and there that aren't related to the task at hand. I just wish offline models were a bit better at Rust.
Sadly, it's not local-first, so data still goes to a cloud. "Architecture: Smart glasses connect to user's phone via BLE; phone connects to backend; backend connects to third-party app servers running the MentraOS SDK" - the project's own instructions for coding agents.
"I really wish I could steer with my coffee mug"
Writing good unit tests is hard.
They got a great deal on introductory offer. www.gsa.gov/about-us/new...
It looks beautiful and intuitive, and I would absolutely be trying to make the logo balanced instead of focusing on just the color palette. Maybe it would help if you could rotate the background without affecting the selected colors?
A screenshot from T-Mobile's Scam Shield screen in their T-Life app showing recent activity of two calls from one number being blocked the first time and allowed through the second time.
After opening the T-Life app on my T-Mobile phone and sifting past their ads to report a scam call, I discovered that the scammers got flagged and blocked as a scam call at first and immediately called back without being flagged. World-class scam blocking foiled by redial.
