noosphr

64 ms·

811.
by noosphr 1y ago
Yes, custom hand crafted model will always outperform general statistical models when given the same compute budget. Given that we've basically saturated the power grid at this point we may have to do the unthinkable and start thinking
812.
by noosphr 1y ago
>This looks like they're not training the single big model but instead have gone off to develop special sub models and attempt to gloss over them with yet another model. That's what you resort to only when doing the end-to-end
813.
by noosphr 1y ago
If only more people knew about tramp-mode.
814.
by noosphr 1y ago
Data that can't leave the premises because it is too sensitive. There is a lot of security theater around cloud pretending to be compliant but if you actually care about security a locked server room is the way to do it.
815.
by noosphr 1y ago
There is nothing simple you can say about the blog in under 1,000 words other than 'neat'. It's also one of a longer series of blog posts about life at the time so you have a good 45min of reading before you can even start sa
816.
by noosphr 1y ago
Deep seek papers are a must to read for anyone who wants to understand how to make LLMs operate at hyper scale. All western labs hide their best results, or at most release summaries that are about as meaningful as the answers Cleo used to
817.
by noosphr 1y ago
From the good old days: https://152334h.github.io/blog/non-determinism-in-gpt-4/ (that's been a short two years). Unless something has fundamentally changed since then (which I've not heard about) all sp
818.
by noosphr 1y ago
Yes, we also covered that jumping off buildings doesn't kill people. The landing does.
819.
by noosphr 1y ago
Given that I've got two people telling me here "ackshually" I guess it may not be hallucinations and just really terrible training data. Up next - ChatGPT does jumping off high buildings kill you? >>No jumping off high
820.
by noosphr 1y ago
Two days ago when my boomer mother in law tried to justify her anti-cancer diet that killed Steve Jobs. On the bright side my partner will be inheriting soon by the looks of it.
821.
by noosphr 1y ago
This is where feeding in extra context matters. Paste in text that shows up from a google search, textbooks preferred, to get in depth answers. No one builds multi shot search tools because they eat tokens like no ones business, but I'
822.
by noosphr 1y ago
>they rarely test and often inform their athletes when a test is coming. the big money basically assures that the incentive to dope is also big. but you'll never get caught if the testing process is a joke, so there is nothing to re
823.
by noosphr 1y ago
Do you know what the most cost effective hardware for general internet stuff is currently? I do ML and have to deal with removing tens of kilowatts from a small closet when it comes to on prem stuff - mainly for compliance reasons - and I h
824.
by noosphr 1y ago
Are there any good papers on it? I've found the majority of crypto literature to be between bullshit and terrible.
825.
by noosphr 1y ago
https://github.com/termux/termux-x11 By default it launches with xfce. The best tutorial for using it is: https://ivonblog.com/en-us/posts/termux-x11/ please ignore the pervy anime.
826.
by noosphr 1y ago
>Bitcoin wasn’t designed for high throughput >>Commerce on the Internet has come to rely almost exclusively on financial institutions serving as trusted third parties to process electronic payments. While the system works well enou
827.
by noosphr 1y ago
Blockchains have clearance rates that are a few dozen orders of magnitude too slow for this. Bitcoin for example clears 7 transactions per second.
828.
by noosphr 1y ago
High level and opinionated. The jvm bytecode is much closer to wasm than Java ever was.
829.
by noosphr 1y ago
Its more useful for python devs since pretty much all ml code is python wrappers around c++.
830.
by noosphr 1y ago
Taking Actually Indians to its natural conclusion by cutting out the middle man API.
831.
by noosphr 1y ago
People act like you need a petaflop for something to be useful. The US census of 1890 would have been impossible to complete before the next one started without puchcard machines. One may as well say the information age arrived then. Anyone
832.
by noosphr 1y ago
The entire population of Norway fits in Queens and Brooklyn. If everyone there decided to whittle spoons we'd be midly concerned about just what got in their water, but it won't be an existential crisis for the rest of us. I will
833.
by noosphr 1y ago
My grandmother wasn't using altavista in 2000. She is using chatgpt in 2025.
834.
by noosphr 1y ago
Why? We've managed to live with color and colour without destroying the space time continuum somehow.
835.
by noosphr 1y ago
Are you planning on using humanoid robots as drop in replacements for humans using the same tools?
836.
by noosphr 1y ago
You're right, but invariants are the most bang for the buck you can get out of formal methods without having to use a DSL built for computer assisted proofs. That they work for binary search is a very strong signal to people not famili
837.
by noosphr 1y ago
It absolutely is, but 99% of programs the average person wants to write for thier job are some variation of, sort these files, filter between value A and B, search inside for string xyz, change string to abc. LLMs are good enough for that.
838.
by noosphr 1y ago
We saw it happen already with Deekseek. SV has already thrown it down the memory hole but for a good three months, until everyone else copied their paper, the SOTA reasoning model available to the public was open source, Communist[0] and ca
839.
by noosphr 1y ago
Again, if I want to import pron from Japan I need to not only prove I'm 18 to the border censors but make sure the pron is legal locally. Plenty of people have been arrested for importing things legal in Japan that are illegal in the W
840.
by noosphr 1y ago
If you need an id to buy porn irl why wouldn't you need one to buy it online?

More ›