@funplings whether you think prettier people are nicer depends a lot on how much you value different axes of kindness.
it also depends on what axes of attractiveness you care about. some reflect the genetic lottery, some confirmation to social norms, some artistic taste, some wealth.
@funplings interesting thought. I think it depends on what you consider nice, and what you consider pretty. being pretty is correlated with all good things in life. a good life makes you nicer (more emotional capacity as you say), but also meaner (less understanding of others struggles)
@cloneofsimo the strength of the open-weight models is mostly in line with the data and compute they have access to; I do not think distillation is a significant factor
@NathanpmYoung@AaronBergman18 I can't speak to the SF system, but for San Mateo County libraries (around $25 per visit), a large fraction of those visits are kids staying for multiple hours after school while their parents are at work, or community educational programs
@ErnestZamelczyk@TravisMWhitaker a frame drop every minute is still frequent enough to annoy the player, so Gen2 pauses are often problematic. this is not a hypothetical, it is very obvious in profiling graphs and player experience
@stimmtdochgarn1@lucasmeijer it is worse average time performance, but that doesn't matter. the point is that spikes specifically are a really big problem, because you drop frames.
@stimmtdochgarn1@lucasmeijer that is literally what reference counting does, it is not uncommon at all. and the cost of a chain of frees is much less then GC, and doesn't block other threads.
@stimmtdochgarn1@lucasmeijer the alternative if you didn't use GC would just be to free memory after you are done using it, which doesn't produce "spikes", its evenly distributed across frames. This is the appeal of not using GC
@usr_bin_roygbiv ๐ซคin my experience, it's pretty common in graphics code, where you literally cannot 'auto-research loop,' because the metric of success is a combination of performance + subjective visual appearance
@usr_bin_roygbiv idk there's still optimizations I do that 5.5 couldn't, and require understanding the code line-for-line. though the common thread with those optimizations is they slightly change behavior, just in ways that are acceptable