hack
@hacksparr0w
Followers
366
Following
8K
Media
175
Statuses
1K
physics dropout, previously @seznam_cz swe, built https://t.co/8XKuaC9k4S, currently working on robotics & ai
🇪🇺
Joined April 2020
stop playing stupid games; there's only one real game and that's the game against nature
2
0
21
it's not the same consciousness humans posses because the evolutionary pressures that gave rise to their brains are vastly different. but do they have a model of themselves that would make it easier to predict the next token? probably. will this model become more detailed and
llms might be conscious, we do not know. anyone who's certain about this one way or another is a fool
1
0
2
seems like i'll be able to spend the next 4 days just coding. let's fucking go. let's see if i can get this Python gRPC reimplementation done.
0
0
0
@eniolaslawn @0xfud0 if you're just starting, you'll have to grind through the basics to get some basic intuition. this is best done by reading books and working through exercises. some resources: - gilber strang's linalg book (and his MIT courses) - stewart's calculus textbook - 3blue1brown's yt
1
2
1
gm. reimplementing google's grpc library because their python support for web-grpc is ass. what are you building?
0
0
0
got some spare money, thinking of building an EU-native inference/tranining GPU cluster. industry-centered workloads from EU startups would be highly prioritized. will probably try building a website to gauge the interest in this kind of idea. any of my EU moots interested?
2
0
3
give me 2000 nvidia H100s and i'll automate 90% of your government workers. -- Archimedes
0
0
0
So longer read of DeepSeek-OCR It’s an engineering achievement. It has been suspected for a while that VLM/OCR models could be significantly smaller. The pre-VLM state of the art, Google Cloud OCR would not be more than a 100m model. More recently, relatively small open weights
27
89
1K
OMG a DeepSeek release just for me. >1 million pages per day per node mean could reprocess all of cultural heritage worldwide for next Common Corpus.
Massively unexpected update from DeepSeek: a powerful, high-compression MoE OCR model. > In production, DeepSeek-OCR can generate 33 million pages of data per day for LLMs/VLMs using 20 nodes (x8 A100-40G). They want ALL the tokens. You're welcome to have some too.
7
21
419
Again to live is to fight entropy The nature of things is nature taking over everything But we kept fighting every waking hour to make sure we have the reality we live in
“To be successful at anything, you don’t have to be special. You just have to be what most people aren’t: consistent, determined and willing to work for it. No shortcuts.” — Tom Brady
0
1
2
in czechia we have been building with autonomous agents for a very long time. this is basically our codex for the physical world.
Amazing story: the Czech government spent six years planning a series of dams. A family of beavers constructed the dams for free, in 1-2 says, in the same locations that human picked, accomplishing the goals set by the Czech government and saving humans $1.2 million
0
0
1
It's funny how everyone immediately forgot that the UK doesn't have iCloud encryption anymore. Never trust Apple, or any BigTech company. They don't care about you.
7
4
157
🇨🇳 China Humanoid Robot company AheafFrom achieves faces with human like expressions with AI, They combine "self-supervised AI algorithms" and wide-range "bionic actuation" to "express authentic emotions and lifelike facial expressions." https://t.co/7rKljTYcQQ
46
142
671
during the previous RL boom, we lacked models that would possess a general understanding of the world. without semantic understanding or common sense, agents blindly explored millions of states, requiring elaborate reward engineering just to learn what any human would grasp
0
0
2