Zhipeng Huang
@nopainkiller
Followers
832
Following
3K
Statuses
8K
Open Source Effectively Accelerates !
33.642931,-117.84131
Joined May 2009
the first blog deals with compression, which largely refers to artificial intelligence. I share eight high level trends that I think the field is heading into post-2025, to make the bioverse compression doable. Read the blog here 3/nš§µ
1
2
9
RT @keenanisalive: I don't know who needs to hear this but: "SDF" has, for some, become a catch-all term that actually means "a function tā¦
0
60
0
@EMostaque probably one of the good governance papers you are looking for
I explain more about these and many other implications of the shift to scaling inference compute in my new paper. Take a look: 16/16
0
0
0
It begins
0
0
1
RT @leloykun: I wanna highlight this! There already is a competitive, speedrun-style benchmark comparing optimizers. It's called AlgoPerf!ā¦
0
1
0
RT @teortaxesTex: on a pettier note, a personal first for a major lab's release: every author here (except Dejian Yang who seems to not havā¦
0
1
0
RT @dileeplearning: Exciting! 'Space is a latent sequence' & CSCG theory of hippocampus we developed at @GoogleDeepmind and @vicariousai geā¦
0
25
0
@georgejrjrjr š¤£ that url. I think it could be broke down to have like each item a project, say 1.C filtered prompt alone would be an awesome project on its own. For the track either the foomyet or have a somewhat leaderboard make sense ?
0
0
1
RT @teortaxesTex: just R1 and basic verifier in a loop. No tools for R1 as such. No finetuning on proprietary Nvidia code. R1 isn't even thā¦
0
38
0
But to put it cruelly, that gap is just one earthquake away...
> While industry observers note that Samsungās 3nm yield might have reached around 30%, it is reportedly still at about 20% in reality. ā¦ 10% for 2nmā¦ Brutal. Even with full access to EUV tools, catching up to TSMC is incredibly hard.
0
0
1
RT @kaepora: Such a great intro book to quantum field theory, thanks so much @DefuseSec for the recommendation!
0
1
0
RT @georgejrjrjr: @teortaxesTex The information bottleneck modern training regimes provide seems sufficient. ie, I don't think this is aā¦
0
1
0
RT @teortaxesTex: @doomslide It's the softwareā¦ for Nvidia hardware design and large parts of it are in Nvidia engineers' wetware
0
1
0
RT @Modular: CUDA is the undisputed king of GPU computeābut why? @clattner_llvm's latest post in our Democratizing AI Compute series breaksā¦
0
19
0
RT @srush_nlp: Talk: we obsess about the cost of training, but often write off data selection, generation, and distillation. In recent workā¦
0
32
0
RT @gammaofzeta: On this (Women and girls in science)day, Iād like to honor Emmy Noether: the most brilliant mathematical physicist ever. Iā¦
0
217
0
RT @KaixuanHuang1: Do LLMs have true generalizable mathematical reasoning capability or are they merely memorizing problem-solving skills?ā¦
0
111
0
RT @cppape: After a long journey, Segment Anything for Microscopy is now published in Nature Methods! We significantly improve SAM for intā¦
0
105
0