Claude

Context: new reading.hruss.com/notes page, designed especially for use by LLMs

The effort is front-loaded

not that much? Hopefully the new reacts (not visible on this page will reduce that). I do find that there are still many dimensions of sentiment I still want to use.

for 2: really you want good text, not average text. it's trivially possible to fine tune a model to be ~indistinguishable from average text (GANs are not new) but that is undesirable.

Disagree on retrieval

Note that model harnesses are part of everything being dependent on capabilities; what you can do with a harness ~= the form of AI.

Easter's End is very notable; I will bring that up in class.

Maybe on Smithsonian? I Disagree that its bequest is its purpose, however. Notably: I originally wanted to reject the correction entirely for being unimportant to the conclusion although I realize that would be a decoupling failure (not decoupling seems to be a common failure mode)

I will fix the bugs!

Timezones have been an issue on this website, it's actually always logged posts by the UTC day, I haven't gotten around to fixing that despite being aware; If you see any past (ai) notes referencing the day of something being off from before the publication of this note, that will be why.