One thing I hate about modern TV shows is that they have been further sliced into ~5-10min sequences between ad breaks, and even if you watch them without ads, you get narratively unnecessary cliff hangers just before a break, complete with dramatic music and a closeup of some dramatic gesture, trivially resolved in the next 5 seconds after the break.
You're constantly yanked out of the narrative in service of ads even if you never see them, which has disfigured the medium.
But it's a continuum, not a hard cutoff. They start hallucinating as soon as you query something they haven't learned verbatim, and they hallucinate/extrapolate sucessfully up to a point, beyond which they start bullshitting, maybe up to a further point where they start saying "I don't know".
The key question is where the boundaries are. Maybe they should be part of the response - a per sentence or per paragraph "confidence scale" that signals how hard they extrapolated from their trained space (I know transformers work per token, but sentence/paragraph would be better human UX).
Of course, if they were trained on garbage input, that would only tell you how accurately they sticked to the garbage. But it would still be invaluable instrumentation for the end user, not to mention for the API provider. They could look at high demand subjects with low confidence answers and prioritize that for further training.
I hate this contant emphasizing with a burning passion; even some newspapers do it. It's like trying to hold a conversation with someone that shouts the important bits to your face.
Head trackers don't map movement 1:1, more like 1:10 (or so). Thus if you want to look 180 back, you turn maybe 30degs but can still eye the screen. Basically you gesture with your neck.
It sounds awkward but it works very well as an input method. It's also way less intense than VR for longer sessions.
Not saying it's absolutely superior, just that it's pretty good as designed.
If you're interested in the subject, the book "Longitude" by Dava Sobel (mentioned in Resources) is a very good read with lots of historical perspective on how these clocks were built.
Have a look at visidata, a TUI table viewer/editor with vim keybindings. It's incredibly powerful and fully automatable - everything you do gets recorded as a series of commands which you can save and replay.
Not the right tool for everything, but it shines for quickly glancing at the shape of tabular data and making sense of it. Sorting, filtering, joins across files, column histograms and even column splitting/rejoining are all keystrokes away.
It groks anything even remotely table shaped like CSV, JSONL, JSON, even Excel files, it can even directly connect to databases and parse tables out of HTML.
Does anyone else feel like there might be a media attack on Uber lately, possibly funded by competitors or opposing municipalities/taxi companies?
It seems like a lot of different kind of bad news have surfaced lately about the company (sexism, expense > income, CEO asshattery, screwing over drivers).
All of these seem legit, but seeing them all blow up over the space of a few months makes me think there might be a concerted effort to hurt the company.
Not saying the company does not deserve it though.
You're constantly yanked out of the narrative in service of ads even if you never see them, which has disfigured the medium.