“we have trained a humanoid transformer with large-scale reinforcement learning in simulation and deployed it to the real world zero-shot” https://learning-humanoid-locomotion.github.io/
EU AI Act: "The legislation ultimately included restrictions for foundation models but gave broad exemptions to “open-source models,” which are developed using code that’s freely available for developers to alter for their own products and tools. The move could benefit open-source AI companies in Europe that lobbied against the law, including France’s Mistral and Germany’s Aleph Alpha, as well as Meta, which released the open-source model LLaMA." https://www.washingtonpost.com/technology/2023/12/08/ai-act-regulation-eu/ [https://archive.is/ZmrC6]
“Sperm whales have equivalents to human vowels. We uncovered spectral properties in whales’ clicks that are recurrent across whales, independent of traditional types, and compositional. We got clues to look into spectral properties from our AI interpretability technique CDEV.” https://osf.io/preprints/osf/285cs
Google’s NotebookLM aims to be the ultimate writing assistant. https://www.wired.com/story/googles-notebooklm-ai-ultimate-writing-assistant/ [https://archive.is/i6Num]
“Nothing lasts forever. Even the universe has several possible endings. Will there be a dramatic Big Rip or a Big Chill–also known as the heat death of the universe–in trillions of years? Or will vacuum decay, which could theoretically happen at any moment, do us in? Perhaps the death of a tiny particle – the proton – will bring about the end.” https://bigpicturescience.org/episodes/end-of-eternity
Breakthrough in 3D printing could enable access to personalised prosthetic devices anywhere in the world https://www.lboro.ac.uk/news-events/news/2023/november/3d-printing-lower-limb-prosthetic-sockets/
New study suggests that attending in-person lectures makes no difference to student achievement. https://doi.org/10.1177/00472395231166592
Political links:
Venezuela, Guyana & The Essequibo Crisis - Posturing or a new Special Military Operation? https://www.youtube.com/watch?v=mWSE9dPEx6Y
Compilation footage from the 110th mechanized brigade on the situation near Avdiivka. A graveyard for heavy equipment. https://twitter.com/NOELreports/status/1733948725205074420
GIF of Russia's offensive slog on Avdiivka after two months. https://twitter.com/georgewbarros/status/1733883074218164632
SBU unit 'A' showing their work against Russian infantry. They report that in a week time, over 500 Russian military men were killed with the help of FPV drones. https://twitter.com/NOELreports/status/1733790453433078043
“BAE systems in the UK will be capable of producing more than 1.5 million large caliber artillery shells annually in the near future. This is a significant increase vs my estimates a few months ago as I have now been able to find the pre war production levels.” https://twitter.com/cameron19460429/status/1733670684700594436
"Germany will be able to deliver ~200.000 artillery shells to Ukraine next year. This is a significant part of the planned million," German Chief of Defence, Carsten Breuer said. He added that German industry is currently increasing production. https://www.faz.net/aktuell/ukraine/generalinspekteur-ueber-zustand-der-bundeswehr-nicht-ausreichend-aufgestellt-19370663.html [https://archive.is/ajmaz]
“Unless the West provides brigades like the 47th with ammunition, they will be unable to stop Putin’s troops.” https://www.thetimes.co.uk/article/surrounded-and-low-on-ammo-the-elite-troops-out-to-spoil-putins-new-year-mdtqkspvj [https://archive.is/h2Sjn]
FYI:
https://www.science.org/doi/10.1126/science.adi8474
Backpropagation-free training of deep physical neural networks
Abstract
Recent successes in deep learning for vision and natural language processing are attributed to larger models but come with energy consumption and scalability issues. Current training of digital deep learning models primarily relies on backpropagation that is unsuitable for physical implementation. Here, we proposed a simple deep neural network architecture augmented by a physical local learning (PhyLL) algorithm, enabling supervised and unsupervised training of deep physical neural networks, without detailed knowledge of the nonlinear physical layer’s properties. We trained diverse wave-based physical neural networks in vowel and image classification experiments, showcasing our approach’s universality. Our method shows advantages over other hardware-aware training schemes by improving training speed, enhancing robustness, and reducing power consumption by eliminating the need for system modelling and thus decreasing digital computation.