Links for 2023-09-20
AlphaMissense, a new AI system from Google DeepMind that can classify whether genetic mutations (missense variants) are benign or harmful - a critical step toward uncovering causes of many diseases, from cystic fibrosis to cancer. https://www.deepmind.com/blog/alphamissense-catalogue-of-genetic-mutations-to-help-pinpoint-the-cause-of-diseases
“We’re taking our most capable model version and allowing you to collaborate with your Google apps & services like Gmail, Docs, Drive, Maps, and more.” https://blog.google/products/bard/google-bard-new-features-update-sept-2023/
From Sparse to Dense: GPT-4 Summarization with Chain of Density Prompting https://arxiv.org/abs/2309.04269
PDFTriage: Question Answering over Long, Structured Documents https://arxiv.org/abs/2309.08872
Clinical Text Summarization: Adapting Large Language Models Can Outperform Human Experts https://arxiv.org/abs/2309.07430
Dr. Oracle medical AI available utilizes published literature AND evaluates their strength when formulating answers. https://www.droracle.ai/
Agents: An Open-source Framework for Autonomous Language Agents — “Agents is user-friendly as it enables non-specialists to build, customize, test, tune, and deploy state-of-the-art autonomous language agents without much coding.” https://arxiv.org/abs/2309.07870
Numenta Unveils NuPIC: Brain-Based AI Platform Promises 100x Efficiency — NuPIC's architecture allows Large Language Models (LLMs) to run efficiently on CPUs https://www.numenta.com/platform/
Open Interpreter: an open-source, free, and unrestricted version of OpenAI’s Code Interpreter that runs locally on your machine. https://github.com/KillianLucas/open-interpreter/
Generative Image Dynamics https://generative-dynamics.github.io/
“each iron atom can catalyze reactions that convert up to 8,000 molecules of carbon dioxide to plankton or algae.” https://www.nytimes.com/2023/09/14/opinion/geoengineering-climate-change-ocean.html [https://archive.ph/7RcrS]
What's spooky about this is it was calculating successfully *several moves ahead*, which I wouldn't expect it to be able to do particularly well.
Source: https://twitter.com/nabeelqu/status/1703969997347856794
Note: The significance here is that it provides more evidence against the hypothesis that large language models simply copy from their training data.
these pattern embeds are going to provide a ton of credence to schizos and absolutely eat us with an incredible density of concept work that could drive the sanest mad by turning infinite credible signals to noise
Source: https://twitter.com/NPCollapse/status/1704169246819791211
More: https://twitter.com/fabianstelzer/status/1704146823877730683
Ukraine links:
The Battle of Bakhmut as seen by a combat medic with Ukraine’s 3rd Assault Brigade https://youtu.be/VCX-xMRvdAU?si=aYmzoPxvo4lLW5Gv
“It appears that Ukrainian forces were able to stop a pair of Russian armored pushes towards Novoiehorivka, Luhansk Oblast in the last few weeks, inflicting heavy casualties.” https://twitter.com/Osinttechnical/status/1704111186780795141
Germany’s new arms package to Ukraine https://twitter.com/deaidua/status/1704154079205699890
Ukraine’s special services ‘likely’ behind strikes on Wagner-backed forces in Sudan https://edition.cnn.com/2023/09/19/africa/ukraine-military-sudan-wagner-cmd-intl/index.html
A Ukrainian MiG-29 fighter aircraft was damaged by a Russian Lancet loitering munition at the airfield of Kryvyi Rih. https://twitter.com/John_A_Ridge/status/1704172340689473984
The moment of the attack on the oil depot in Sochi, as a result of which the tank caught fire. https://twitter.com/NOELreports/status/1704388339585978618