From UC berkeley, stanford, carnegie mellon, & google deepmind.
Octo.
An Open-Source Generalist Robot Policy.
Large policies pretrained on diverse robot datasets have the potential to transform robotic learning: instead of training new policies from scratch,…
From 13 billion light-years across the gulf of space and time, we’ve just caught a glimpse of the most distant black hole merger discovered yet.
Using JWST, an international team of astronomers has discovered two supermassive black holes, and their attendant galaxies, coming together in a colossal cosmic collision, just 740 million years after the Big Bang.
This discovery could be a clue that helps us piece together where supermassive black holes came from, and how they grew so large, so early in the history of the Universe.
From Microsoft.
Towards Modular LLMs by Building and Reusing a Library of LoRAs https://huggingface.co/papers/2405.
The growing number of parameter-efficient adaptations of a base large language model (LLM) calls for studying whether we can reuse such trained adapters to improve performance for…
Join the discussion on this paper page.
The International Energy Agency predicts that artificial intelligence will consume 10 times as much energy in 2026 as it did in 2023.
Optical neural networks, which use photons instead of electrons, have advantages over traditional systems. They also face major obstacles.
Black holes are intriguing astronomical objects that have a gravitational pull so strong that it prevents any object and even light from escaping. While black holes have been the topic of numerous astrophysical studies, their origins and underlying physics remain largely a mystery.
Scientists have discovered that a “single atomic defect” in a layered 2D material can hold onto quantum information for microseconds at room temperature, underscoring the potential of 2D materials in advancing quantum technologies.
The defect, found by researchers from the Universities of Manchester and Cambridge using a thin material called hexagonal boron nitride (hBN), demonstrates spin coherence—a property where an electronic spin can retain quantum information —under ambient conditions. They also found that these spins can be controlled with light.
Up until now, only a few solid-state materials have been able to do this, marking a significant step forward in quantum technologies.
The NA64 experiment started operations at CERN’s SPS North Area in 2016. Its aim is to search for unknown particles from a hypothetical “dark sector.” For these searches, NA64 directs an electron beam onto a fixed target. Researchers then look for unknown dark sector particles produced by collisions between the beam’s electrons and the target’s atomic nuclei.
The future of optical communications just got brighter. In a development reported in Advanced Photonics, researchers from Nanjing University have introduced iso-propagation vortices (IPVs), a novel concept that offers a solution to a long-standing challenge faced by scientists and engineers: how to increase information processing capacity while overcoming the limitations of traditional vortex beams.
Spectroscopy is the study of how matter absorbs and emits light and other radiation. It allows scientists to study the structure of atoms and molecules, including the energy levels of their electrons. Classical optical spectroscopy relies on the way particles of light called photons interact with matter. These classical spectroscopy techniques include one-photon absorption (OPA) and two-photon absorption (TPA).
From the rain drops rolling down your window, to the fluid running through a COVID rapid test, we cannot go a day without observing the world of fluid dynamics. Naturally, how liquids traverse across, and through, surfaces is a heavily researched subject, where new discoveries can have profound effects in the fields of energy conversion technology, electronics cooling, biosensors, and micro-/nano-fabrications.
