rio-popper

6 ms·

1. 1.▲
   by rio-popper 9mo ago

   We actually did try making a pulley-system to reduce weight at one point. The issue was that it moved the headset kind of oddly WRT the person's head, which reduced sensor contact. One thing we intend to try is pulley system + good chi
2. 2.▲
   by rio-popper 9mo ago

   For the record, it was two basements -- we moved office in the middle -- and a bigger issue was actually overheating. But your point is basically right! The model is a lot better at certain kinds of ideas than others. Particularly concernin
3. 3.▲
   by rio-popper 9mo ago

   Good point. We're a team of 7 right now (3 engineering, 4 running data collection across shifts). We've been spending ~all our time on the data and model side, so the “About us” page lagged behind, but we’ll add one this week. App
4. 4.▲
   by rio-popper 9mo ago

   so the neural quality real-time checking was the most important thing here. Before we rewrote the backend, between 58-64% of participant hours were actually usable data. Now, it's between 90-95% If you mean the text quality scoring sys
5. 5.▲
   by rio-popper 9mo ago

   Yeah we did consider this. For now, there's an advantage to having the data collection in the same building as the whole eng team, but once we hire a couple more engs, I expect we'll just replicate the collection setup in other co
6. 6.▲
   by rio-popper 9mo ago

   edit: oops sorry misread - the neural data is tokenised by our embedding model. the number of tokens per second of neural data varies and depends on the information content.