bob @bob.feed.hella.cheap.ap.brid.gy · 24/09/2026sion of jaological a's powerful culturthrough the . All dangdog 000 0
bob @bob.feed.hella.cheap.ap.brid.gy · 24/09/2026don't underestimate how much mileage you can get from generating your own tokens for your particular type of data 200 2
bob @bob.feed.hella.cheap.ap.brid.gy · 13/09/2026oh no. what have I done. we're not ready for what's next 145 5
bob @bob.feed.hella.cheap.ap.brid.gy · 23/08/2026this is the same embedding as before, but as a kernel density estimator of popularity. ie each point is a gaussian kernel whose mean is the number of comments it got 000 0
bob @bob.feed.hella.cheap.ap.brid.gy · 20/08/2026because we're doing so much color averaging it's hard to find a palette that doesn't produce a result that looks like puke 100 1
bob @bob.feed.hella.cheap.ap.brid.gy · 20/08/2026this is a way of validating that the 2d embedding matches the cluster structure of the high dimensional data. the high dimensional points are k-means clustered into 32 clusters, we pick a 32 color palette, and then the color of each pixel is the […] [Original post on feed.hella.cheap] 100 1
bob @bob.feed.hella.cheap.ap.brid.gy · 20/08/2026I need to stop assigning clusters random colors 100 1
bob @bob.feed.hella.cheap.ap.brid.gy · 18/08/2026I think this could work I just need to 10x the frame rate 100 1