deleted by creator

Why would you expect that?

Meh. They’re dealing with two orthogonal problems here:

1. There’s a right amount of documentation. It’s not zero, but it’s also not “every thought you’ve ever had”. The more your documentation can be generated or validated automatically, the more you can reasonably sustain.
2. It’s expensive to be wrong. You can deal with that by either doing pre-work in order to reduce the chance you’ll be wrong (which increases the cost of later finding out you were wrong), or sprinting towards a deliverable in order to minimize the cost of being wrong (which increases the chance that deliverable will be wrong).

You’re probably better off looking at each of those problems independently first and deciding where on the spectrum your team would thrive. RFCs might hit the sweet spot for both. But if you don’t ask the deeper question, you might just make things worse.

if you want to, say, stop LLM server if available mem is under 8GB and start it again when it’s over whatever LLM needs (in my case it’s 64GB):

So this is the guy that bought all the RAM.

Why is that image giving the same vibes as:

All of the needed words. Gotta be careful here. LLMs don’t deal in information. But yes, they are good at stringing together tokens if that’s all you need.

It’s pretty apt, honestly. It’s just the next step of the climate-denial and cancer-denial playbooks.

We know that the tech bosses are aware of how harmful their stuff is. We know that they hire experts specifically to make their stuff as addictive as possible. We know they bribe the hell out of politicians to avoid getting regulated. We know they cook their books and launder money like crazy. We know their financial models are predicated on getting everyone to use an ever-increasing dose of their stuff. We know that people suffer horrific conditions to help build their devices and moderate their content cesspools.

It may seem crass to compare tech bosses to narco kingpins. But that’s because their methods are crass. They want to seem sophisticated and unique. But they’re not.

Using an LLM to work with facts is like using JPEG to store x-ray images. Utterly reckless.

It’s stabs all the way down

Eh, it’s fine. It has some bad choices baked into it, but what language doesn’t? And JS in 2025 is miles better than JS in 2005.

I wouldn’t choose it for every project, but it’s a reasonable choice in many cases.

And unlike a regular Linux distro, you’ll have zero leftover systemd units or config files floating around in your FHS dirs. (You’ll have the binaries for Gnome sitting in /nix/store until you do a GC, so you can still quickly switch back if you want to.)

The trunk metaphor doesn’t work very well in Git, because branches aren’t long-lived containers of sequential commits the way they are in SVN. There is no “root” commit that is guaranteed to have a consistent relationship with main, because branches are just names attached to a commit and can be reassigned at will.

To represent 4 values, you only need 2 bits

• 00
• 01
• 10
• 11

More buffer for the git pushin’

deleted by creator

“We have made a commitment to our employees that all of these efficiency gains, and especially the applications of AI, should also, to some degree, come back in their pay cheques so that they are fully … incentivised [and] aligned with the investors, to drive these changes through the company.”

All… to some degree

Also worth noting that you need to be on US soil in order to apply for asylum. It’s common for migrants to cross the border without permission and immediately turn themselves in, because it’s the only reliable way to initiate an asylum review.

If they are later granted asylum, would you still consider that illegal?

I hate this timeline so much