Links for 2023-03-09
How can robots learn manipulation *just* by watching videos of humans in different unstructured settings? A simple framework for enabling zero-shot coarse robot manipulation just from diverse passive human videos. https://sites.google.com/view/human-0shot-robot/
“6 technologies are in the works and will very likely mature within the next 3 years…Soon enough AI-empowered products will be accurate, informative, up-to-date, and efficient.” https://www.codium.ai/post/6-important-ai-near-future-breakthroughs-or-why-the-ai-hype-peak-is-likely-to-be-ahead-of-us
First-principles on AI scaling — “There is no apparent barrier to LLMs continuing to improve substantially from where they are now. More data and compute should make them better, and it looks feasible to make datasets ~10x bigger and to buy ~100x more compute.” https://dynomight.net/scaling/
“Prismer achieves fine-tuned and few-shot learning vision-language reasoning performance which is competitive with current state-of-the-arts, whilst requiring up to two orders of magnitude less training data.” https://shikun.io/projects/prismer
NVIDIA’s New AI: Wow, 30X Faster Than Stable Diffusion! https://youtu.be/qnHbGXmGJCM
ChatGPT's API is So Good and Cheap, It Makes Most Text Generating AI Obsolete https://minimaxir.com/2023/03/new-chatgpt-overlord/
Evidence for the scaling hypothesis continues to accumulate: “ViT-22B demonstrates increasing performance with scale. We further observe other interesting benefits of scale, including an improved tradeoff between fairness and performance, state-of-the-art alignment to human visual perception in terms of shape/texture bias, and improved robustness.” https://arxiv.org/abs/2302.05442
ChatGPT broke the EU plan to regulate AI: Europe’s original plan to bring AI under control is no match for the technology’s new, shiny chatbot application. https://www.politico.eu/article/eu-plan-regulate-chatgpt-openai-artificial-intelligence-act/
Room-Temperature Superconductivity Claimed https://spectrum.ieee.org/room-temperature-superconductor
Scientists create mice with two fathers after making eggs from male cells https://www.theguardian.com/science/2023/mar/08/scientists-create-mice-with-two-fathers-after-making-eggs-from-male-cells
There is a worrying amount of fraud in medical research: “In 2009 plos one published a round-up of 18 surveys of scientists, mostly in America, that had asked about fraud. Though only 2% of respondents admitted falsifying data themselves, 14% said they knew someone who had. Similarly, a third of those asked confessed to other questionable research practices, such as dropping inconvenient data points based on “gut feeling”, or making important changes to a study’s protocol while it was in progress. But they pointed the finger at 72% of their colleagues.” [The Economist] https://archive.is/XFu02
“Nuclear Winter Theory is Wrong as It Assumes Super Flammable Cities” https://www.nextbigfuture.com/2023/02/nuclear-winter-theory-is-wrong-as-it-assumes-super-flammable-cities.html
"[W]omen rated a weak man who was willing to protect them as far more desirable than a strong man who was unwilling to protect them." https://robkhenderson.substack.com/p/an-instinct-thats-basic-and-deeper
MIT Breakthrough: How To Efficiently Remove Carbon Dioxide From the Ocean https://scitechdaily.com/mit-breakthrough-how-to-efficiently-remove-carbon-dioxide-from-the-ocean/
Baby this is gonna blow them all away. It's a neural net processor.