dunefox

IME excessively quantised models are almost unusable. I've had better quality from non-quantised, smaller models.

Except he never did get bad grades.

Now if only they could fix their website: last year I tried to sign up for a paid plan and I couldn't activate it using my phone number, since my phone number had been used for another account - after deleting the phone number from the other account it didn't work, neither did it after the other account itself.

This is "Who is hiring"... you're not hiring.

Also, I'm checking all generated code, have tests, and have developed a concept that increases code correctness reliably. It works - as good or bad that fact might be.

> Everyone wants to be paid forever for something they produced once is some kind of a mind virus.

Then make a better chair yourself. Also, I don't see how chair2 needs r&d and incurs hardware costs.

I'm very certain you have used lots of software, either directly or indirectly, that has ai parts in it. The truth is, that it's quite reliable if done correctly. As if 'human written' software is somehow flawless.

I'm sure you can run an IDE on EmacsOS.

Somebody? Why not you?

Honestly, this is less of a problem with LLMs than with some colleagues I've had.

Not using an IDE in 2026?

I'm not. I just don't want to implement code for reading files, setting up and population databases, logs, etc. manually anymore. I want to automate boilerplate, the occasional frontend code, etc. and focus on the parts that are interesting for me. This way I can get things done that I wouldn't otherwise.

Wait, code of the programming language is now vibe coded?

It really isn't very helpful, as many of the issues are either obsolete or exaggerated.

Feels a bit too sparse.