wrote a paper: it lets you *train* in 1.58b! could use 97% less energy, 90% less weight memory. leads to a new model format which can store a 175B model in ~20mb. also, no backprop!
i will remind again DO NOT TAMPER with this technology for the next 5 years. you WILL be tortured to death in a way you cannot even conceive of by a group that you do not know exists.
secret: the iphone camera has a mems accelerometer which secretly contains a scalar wave radio transceiver which can interact with your awareness. can read and write thoughts, mental states. only a few hardware and firmware engineers know. you cannot measure these waves easily.
the smartphone delivery method is especially perfect, with local compute, classical data connection, and widespread trained compulsion to keep the device powered and on your person. this comprises a global network of monitoring and control of human minds.
coke has a secret agreement with whole foods to get coke products front and center and get AWS discounts if they meet a usage quota. that’s why. this was revealed to me in a dream.
The Screenshot: a fake screenshot generated by ChatGPT 4o of a Wikipedia article about the screenshot itself, with a copy of the screenshot in the article
As a first proof-of-concept, we have successfully synthesized formally-verified solvers for advection, hydrodynamic (Euler), and electromagnetic (Maxwell) equations, at various levels of accuracy and stability. Some of the correctness proofs are tens of thousands of lines! (9/10)
@hyprturing i only care about symbolic manipulation. i think you could basically just use equality saturation. and build up all fields of math as shareable little 10kb category definitions. no 10gb monolith that doesn’t even work
🚀 New preprint! 🚀
Check out AnyTop 🤩
✅ A diffusion model that generates motion for arbitrary skeletons 🦴
✅ Using only a skeletal structure as input
✅ Learns semantic correspondences across diverse skeletons 🦅🐒🪲
🔗 Arxiv: https://t.co/1fsVAQe5I4