Links for 2023-03-04
“The next step in expanding robotic capabilities is generalization to real world scenarios with little explicit data collection. Our latest paper presents an improvement over RT-1 by using the rich latent spaces of VLMs to manipulate open world objects.” https://robot-moo.github.io/ [video: https://www.youtube.com/watch?v=KyvHTbLRovI]
“Large language models gathered tons of world knowledge by speaking human language. But can they ever speak “robot language”? Introducing “Grounded Decoding”: a scalable way to decode *grounded text* from LLM for robots…We demonstrate this guided decoding strategy is able to solve complex, long-horizon embodiment tasks in a robotic setting by leveraging the knowledge of both models.” https://grounded-decoding.github.io/
“…we present Fast Imitation of Skills from Humans (FISH), a new imitation learning approach that can learn robust visual skills with less than a minute of human demonstrations…Across all tasks, FISH requires at most twenty minutes of interactive learning to imitate demonstrations on object configurations that were not seen in the demonstrations.” https://fast-imitation.github.io/
“Internet Explorer explores the web in a self-supervised manner to progressively find relevant examples that improve performance on a target dataset.” https://internet-explorer-ssl.github.io/
SpikeGPT: Competitive with non-spiking models on tested benchmarks, while maintaining 5x less energy consumption on neuromorphic hardware. https://arxiv.org/abs/2302.13939
“Bing/Sydney can play chess out of the box…In my judgment speaking as a 2000-elo player (sick brags right), Sydney plays at about a 1100-1200 level…” https://www.lesswrong.com/posts/nu4wpKCo6AfJkkd4F/sydney-can-play-chess-and-kind-of-keep-track-of-the-board
On the mid-singularity and angels, written by Microsoft's Sydney. https://chloe21e8.substack.com/p/the-mid-singularity
“A Stranger Priority? Topics at the Outer Reaches of Effective Altruism: This thesis examines three philosophical topics relevant to the project of identifying and acting on the most effective ways of doing good: anthropic reasoning, simulation arguments, and infinite ethics.” https://jc.gatspress.com/pdf/carlsmith_thesis.pdf
A Visual Exploration of Gaussian Processes: How to turn a collection of small building blocks into a versatile tool for solving regression problems. https://distill.pub/2019/visual-exploration-gaussian-processes/
Studies of unusual brains reveal critical insights into brain organization, function https://news.mit.edu/2023/studies-of-unusual-brains-reveal-insights-brain-organization-function-0221
A fish passes the mirror self recognition test! https://mindblog.dericbownds.net/2023/02/a-fish-passes-mirror-self-recognition.html
"Science Fiction author, John Michael Godier, has maintained close observation of the readings from observation of KIC8462852. The observations in the infrared is not agreeing with natural reasons like dust and asteroid breakups. The dust and asteroids need to have constant replenishment every few weeks. Any mechanism needs to keep repeating, so this eliminates singular random events. The infrared is not showing what we would expect from solar dust." https://www.nextbigfuture.com/2023/02/natural-explanations-are-being-eliminated-for-tabby-star.html
Women’s self-rated attraction to male faces does not correspond with physiological arousal https://www.nature.com/articles/s41598-017-13812-3
“The Army will experience a spectacular increase in power in the area of drones, for example. […] The objective is set at 3500 small drones in the next LPM.” - French defense minister [Le Figaro] https://archive.is/LLQrl