FrameworkFred

I suspect it's just expensive to do a good job of it.

Amazon used to be better at targeted advertising than it is now. I used to browse for a few things on Amazon, then go do something else, content with the knowledge that they'll do the hard work of figuring out what is actually a good recommendation for it. For a few years, this strategy helped me buy tools, light fixtures, and other things I used around the farm.

I'm not sure when it stopped, but it definitely did.

Ultimately, it's probably just a lot more compute than it's worth. They can probably spend $1 in compute to drive $10 in sales with dumb ads or they can spend $2 to drive $15 with smart ads. It's a diminishing returns thing and they drew the line such that I get dumb ads. Maybe folks that buy more, higher priced items get better ads.

I have a $20/mo Claude subscription for personal projects and a $100/mo Claude subscription for work.

I run Arch Linux on my workstation. I originally ran Claude only in a VM, but have since "sandboxed" by simply adding additional users to my machine and running Claude in those user accounts. I use waypipe to run gui apps for those users on different KDE/Plasma activities so they remain somewhat isolated while I can run the various IDEs, Obsidian apps, etc. in their environments.

These days, I'm just running Opus 5.5 in Claude Code. I've tried OpenClaw and Hermes and didn't get anything interesting to happen. I tried OpenCode with a local Qwen model and it was ok, but slow.

I now run everything from Obsidian. I have QuickAdd commands for creating a new plan prompt, a plan workflow skill that builds the plan under the plan prompt in a subpage, with a handoff doc that's a subpage of the plan. I've got QuickAdd commands for "Send to Claude Session" that opens a tmux session keyed off the plan name so I can run separate sessions per-plan.

The plan workflow also maintains a global "Efforts" page that shows all active and completed plans. The former includes details like open PRs, % complete, next step, etc. and a color-coded button to show whether the session is executing, idle, or closed.

I maintain separate user accounts and obsidian vaults for each major project (work, personal project A, personal project B, etc.).

I had a fujitsu lifebook that was 10 years old when I bought it and it lasted another 8 or so after that. I'd maxxed out the ram to 8GB, replaced the missing stylus and bought a new battery. It had a fantastic display that was brighter and had a wider viewing angle than the system76 gazelle I replaced it with. And the swivel hinge connector was soo solid.

I now have a framework 13 that I'd rate slightly higher than that fujitsu (it's lighter and the changeable ports thing is a truly wonderful idea), but it's a bit of a toss-up.

I wonder if the google ai team has to do some RF to scrub the whole "Don't be evil" thing from its all-written-human-knowledge weights before they can get it to attack.

When I mention a preference for brevity in a comment offhand and it snowballs into a rule 8 sessions down the road in a PR review that requests the removal of 90% of a colleague's docstrings.

I can't imagine all the "push back" on the "load bearing" Code of Conduct from their corporate foundation documents.

I've been in the room when an org who tried to convince law enforcement to go after a human for similar things. It's not easy. Probably won't happen. So, you know, felony "lite".

That was a fun read. I was a draftsman as a teenager in the late 80s and I did my best to make my handwriting look like those letters. Then, as a CAD operator, I put that font on everything I produced. It never once occurred to me that it looked like that because it was designed to be legibly engraved.

TIL I can just view markdown in okular directly instead of using a script that converts markdown to html, then converts html to pdf, then opens that pdf in okular.

that is really cool. well done!

actually, that sounds like a pretty good feedback mechanism, especially if you come up with a way to get lat/long and/or an accurate timestamp.

I derailed after this sentence. I searched for more uses of "rats" in the article, then looked in the HN comments to see if it was a bit of jargon that I was unaware of. I read "Lessons from the art of juggling" years ago, but despite the primer, I couldn't get past the unanswerable question of whether I was, in fact, a "rat".

so far, I've used it to kill a bunch of time trying to get it to respond to "Hi @Kirk" in a private Slack channel.

...and to laugh a little every time it calls me "commander" or asks "What's the next mission?" or (and this is the best one) it uses the catchphrase I gave it which is "it's probably fine" (and it uses it entirely appropriately...I think there must have been a lot of sarcasm in qwen 3.5's training data)

and I've treated it like it's already been compromised the whole time.

It would make sense that the holes were a convenient way of thinking and speaking about large quantities of goods such that tribes of people might want to exchange. It would be a very visual way of comparing dissimilar goods, like "1 hole has 50 alpaca skins and I need 200 for the shelter I'm planning to build, so I need 4" and "1 hole has 8 baskets of dried fish which can last 3 families thru the winter, so I need 3 for the nine families on the farm", etc.

And I bet they paid a bit of rent for the privilege. Pretty cool.

This all makes a lot of sense. It is worth pointing out that AI isn't terrible at cleaning up tech debt. I've absolutely used it cleanup code sprawl and to correct design missteps. It gets a lot of well-deserved blame for generating way too much code that's a lot more complicated than it needs to be, but probably not enough credit for making a lot of the cleanup effort around that sort of code a lot more tractable.

I've been using the "Zim desktop wiki" like this for years. I do recommend it as well...super handy to be able to go looking for my thoughts or snippets from 6 months ago. I can also use git to sync between my desktop and laptop because it's all text.

I agree with this soo much.

In a very real way, a codebase is a conversation among developers. There's every chance the next guy will be less familiar with the code than you, who just did a deep dive and arrived at some critical, poorly documented line of code. There's an even better chance that person will be you, 6 months from now.

Use opportunities to communicate.