retinaros

4 ms·

1.
by retinaros 10d ago
do u live in russia?
2.
by retinaros 11d ago
I knew before clicking it will be one of the things they say. people are sheeps. they will say whatever they think is right up to a limit. then they will break this limit and punish everyone that is deviating from it. this is what happened
3.
by retinaros 23d ago
Selling pro accounts for storage. Its at best a few ten millions arr maybe 100-150
4.
by retinaros 23d ago
Its a s3 wrapper. Can be replaced by a wordpress with torrents too. But they built an ecosystem around it and some libs/content. i guess thats what nvidia value
5.
by retinaros 23d ago
Wasn’t official
6.
by retinaros 23d ago
Life changing for engineers having joined a few year ago. Congrats
7.
by retinaros 1mo ago
somehow I cant understand how luna is a step up. to me it feel dumber than 4.1. slower too
8.
by retinaros 1mo ago
its always the most virtue signaling people (sam on "openess", dario on pretty much every word coming out of his mouth, HF) that ends up doing the worst things because they always tend to extreme. Funny to see them joining the big
9.
by retinaros 1mo ago
I did that on SVG mostly to teach it to draw pelicans but also to generalize it. most of the behavior is from SFT on the base model tho. RL is very ineficient at style or at least at generating novelty out of distrib.
10.
by retinaros 1mo ago
I said there is no evidence distillation is all you need as per the initial comment "The fact that AI models can be so easily distilled and replicated is such a stroke of luck." it is far from trivial to reproduce the capabilities
11.
by retinaros 1mo ago
the initial post infers that distillation is all you need. it is not. in 2026 you need large scale distributed inference of gigantic models, rl envs and millions of dollars runs to get to something decent. if you think glm just has to sft o
12.
by retinaros 1mo ago
Nothing against what I said in there. Distillation is far from enough to get to the frontier. Its at best a ramp (the most efficient one used by everyone ) that shortcut and saves millions of rl runs before a model moves.
13.
by retinaros 1mo ago
there is no evidence. it shortcuts post training by a huge margin this is true. but that is all.
14.
by retinaros 1mo ago
at this stage europe is a machine for politicians to become wealthy through corruption, mass immigration, censorship, mass media control. that is why they hate X and are now fighting to ban it.
15.
by retinaros 1mo ago
I cant read it. Sounds like opus slop. Should have used glm to write it
16.
by retinaros 1mo ago
Thing is they chose macron. They literally voted for him to remove from working people the same right they benefited. I do t inow why its hard for you To grasp. The data is public.
17.
by retinaros 1mo ago
i am actually a french muslim immigrant and all my life has been people like you - that likely identify as left and good willing people - telling me what to do and telling me i am indignant because I am uneducated or I dont follow what the
18.
by retinaros 1mo ago
yeah. their fraternité ends when they need to fairly contribute.
19.
by retinaros 1mo ago
what bigotry? they are stealing the money and future of every french child being born and you defend them insulting people showing you the facts of bigots.
20.
by retinaros 1mo ago
there was another leak recently and full data leaked. people having crypto got victim of home invasion , beatings, kidnapings, so much that france accounted for 70% of all crypto crimes the past years. and It wasnt for millions. people even
21.
by retinaros 1mo ago
france is living through its own programmed obsolescence and it is 100% due to boomer generation locking in the vote to favor themselves instead of france future. can you believe that during covid governement pushed to raise retirements pen
22.
by retinaros 2mo ago
its not exactly the same... its tool use spec asking to put thinking in inputs fields... it is a good idea but its not same.
23.
by retinaros 2mo ago
curious seeing how anthropic is agressively fighting this stuff how did you get to experiment on this? did you just try and shown them results or did you need approval first? I am interested mostly because I research on distillation
24.
by retinaros 2mo ago
many are buying into routing as a way to save money but it destroys the cache savings. prob that is what copuilot is doing
25.
by retinaros 2mo ago
the article means code as a whole not just the moment you input some if else in a screen but the whole act from thinking about it to make it live in prod.
26.
by retinaros 2mo ago
any link to this caching tech?
27.
by retinaros 2mo ago
https://x.com/Hesamation/status/2082078307147776117?s=20
28.
by retinaros 2mo ago
it cant be explained. but distillation shortcut training time by a lot. you need much more than that to be near SOTA obviously
29.
by retinaros 2mo ago
aws its their business to make ur data safe. frontier lab business is to use your data and mine to train models
30.
by retinaros 2mo ago
on 1 paralel request maybe

More ›