Well, the home folder was encrypted, and the hard shutdown had borked the headers in such a way that the decryption was failing. I suppose a few hundred hours of technical analysis might have retrieved the files, but luckily it was a new PC and I only had about 100 hours of work on it to start with.

OS-X (they still use that, right? Not iOS desktop or somesuch nonsense, yet?) seemed pretty much a middle ground between Windows and Linux the last time I used it. Kinda slightly more polished and uniform presentation than Ubuntu-du-jour, a little less mysterious than Windows, but in the end: just as screwed up.

I tried enabling Home folder encryption. After about 3 days a hard power-off shutdown (needed due to a driver error in their walled-garden hardware MacBook Pro, it wouldn’t power off or restart any other way) then the encrypted home folder was toast, unretrievable - laptop wouldn’t boot. Tech support was very nice, reassuring that they knew what was going on, and their best solution? Reinstall the OS from physical media, start over fresh, your files are so secure that not you or anybody else on the planet will ever see them again.

People should really consider if the mental models they have in their heads about different operating systems are actually based on reality.

These people’s reality is: they are familiar with Windows, and anything else is scary and perceived as even more difficult to learn to use. 20 years ago a colleague asked me about changing to Linux, I told him he could do all the same things he was doing, just use Open Office instead of MS Word and Excel, GIMP instead of Photoshop - he didn’t even dive as deep as the differences between GIMP and Photoshop usage, his response was: “You mean I’ll have to learn all new icons and names for my software?” “Well, yeah, that’s part of moving.” “In that case I don’t think Linux is for me.” “I have to agree with you there.”

The very interesting part will be how successful they are at training the training data selectors to choose high quality data sources.

I feel that a lot of what is improving in the recent batch of model releases is the vetting of their training data - basically the opposite of model collapse.

Nothing requires an LLM to train on the entire internet.

It’s overhyped in many areas, but it is undeniably improving. The real question is: will it “snowball” by improving itself in a positive feedback loop? If it does, how much snow covered slope is in front of it for it to roll down?