pull down to refresh

100 sats \ 7 replies \ @carter OP 22 Aug \ parent \ on: Making LLMs Cheaper and Better via Performance-Efficiency Optimized Routing AI

I wonder if its related to this phenomenon where running 2 times gives much better results https://brooker.co.za/blog/2012/01/17/two-random.html

write
preview
0 sats \ 6 replies \ @optimism 22 Aug

I read that as: when you gamble twice, you have a greater chance to win once. lol

reply
100 sats \ 5 replies \ @carter OP 22 Aug

I think the big thing is this: Just random is the same no matter the delay, picking the fastest of 2 is much better but not that much worse than picking from 3. It seems similar to this strategy https://www.tiktok.com/t/ZP8BD7XQp/

reply
100 sats \ 4 replies \ @optimism 22 Aug

bro made me watch tiktok!

It's a good theory, but the reason why I say gamble is because of the randomization going on, even in MoE, where it's been reduced a lot. I'm not sure how this works in gpt-5 or claude-4 though, so maybe that's worth testing too.

reply
100 sats \ 3 replies \ @carter OP 25 Aug

I wish it would download and embed them like it does for twitter

reply
56 sats \ 2 replies \ @optimism 25 Aug

I'd settle for the mp4 as upload being embedded...

reply
100 sats \ 1 reply \ @carter OP 25 Aug

yeah... not always downloadable

reply
35 sats \ 0 replies \ @optimism 25 Aug

yt-dlp

reply