NASA - National Aeronautics and Space Administration’s Post

🐝 🔭 IBM and NASA built an open-source scientific LLM tl;dr: IBM and NASA have developed advanced language models to simplify access to scientific knowledge. By utilizing a custom tokenizer for scientific terms, they outperform existing models in benchmarks, making complex scientific information more accessible and understandable for all. * * * 🤖 IBM and NASA - National Aeronautics and Space Administration have joined forces to build a suite of efficient language models, trained meticulously on an expansive corpus of scientific literature spanning astrophysics to biological sciences. These aren't your run-of-the-mill models; they're capable of tasks as varied as entity extraction to answering the questions that push our understanding of the universe forward. What sets these models apart is their unparalleled performance, outshining others by up to 5% on benchmarks that matter to scientists and researchers. And IBM and NASA have made these tools available on Hugging Face, inviting all to explore further than ever before. 🚀 This collaboration isn't just about advancing science; it's about democratizing access to knowledge, and the fact that heavyweights like IBM and NASA are behind it gives us hope that we'll see more non-commercial, high-performance domain-specific LLMs in the future. #Innovation #IBM #NASA #LLM #AI

🤓🤯 The sheer volume of potential applications for this new #supermodel simply blows my mind! Yet another cool #foundation from the NASA - National Aeronautics and Space Administration IBM colab 🤗 “For the benefit of the scientific community, the NASA - National Aeronautics and Space Administration IBM team has released the developed models and will release the benchmark datasets that span named entity recognition for #climatechange, extractive QA for #Earth science, and information retrieval for multiple domains. The #INDUS encoder models are adaptable for science domain applications, and the INDUS retriever models support information retrieval in RAG applications…” A link to the INDUS paper https://lnkd.in/eSu8URqq #foundation #frontiers #sciencedomain #generativediscovery

Icosa Computing has just released its first academic paper in collaboration with NASA - National Aeronautics and Space Administration, USRA , and Hewlett Packard Enterprise. 🌏 👨🚀 ⭐ Their research highlights a significant advancement in Large Language Models (LLMs). While it's well-known that the choice of prompting techniques can greatly impact LLM performance across various tasks, the most effective techniques typically require human-crafted prompts tailored to specific tasks. 🧠 Introducing Combinatorial Reasoning (CR): Icosa Computing's innovative framework for fully-automated prompting. CR samples reasons from an LLM pipeline and transforms them into a Quadratic Unconstrained Binary Optimization (QUBO) problem, optimizing prompt selection and enhancing performance without the need for manual intervention. PS. I included the figure below because it's cool. 😅 For more details, please find Mert Esencan or the Icosa Computing team. 👏 💜 https://lnkd.in/dgD2krA6 #CombinatorialReasoning #LLM #IcosaComputing

#Nasa and #IBMResearch published a paper on the #LLM for #science work that we did jointly. The author list of 34 spans 3 labs of IBM Research, 5 divisions of NASA, 2 university related entities (#UAH & #Harvard-#Smithsonian CfA) and one other firm (#Navteca). The paper covers the encoders and sentence transformer models that was developed and open source. Also includes 3 benchmarks in IR, NER and QA which have also been open sourced. The experimental section has comparisons against state of the art general purpose as well as science related models. The models have already been downloaded 1000+ times Attached is the link to the paper https://lnkd.in/ehCTCv5V #deeplearning #artificialintelligence #ai #ibm #science Rahul Ramachandran Manil Maskey Muthukumaran R Iksha Gurung Aashka Trivedi Masayasu Muraoka Takuma Udagawa Bharath Dandala, PhD Kaylin Bugbee Lauren Sanders Sylvain Costes Peter W. J. Staar Armin Mehrabian Megan Ansdell Alberto Accomazzi Sanaz Vahidinia Cesar Berrospi Michele Dolfi, PhD Felix Grezes Ryan McGranaghan Davis Wertheimer S. Karthik Mukkavilli Yousef El-Kurdi Tsengdar Lee

The Laboratory for Innovation Science at Harvard is redefining innovation.  💡 Partnering with organizations like Harvard Medical School, Broad Institute of MIT and Harvard, NASA - National Aeronautics and Space Administration, and Procter & Gamble, they use field-based experiments to explore open innovation, the science of science, and the impact of Data Science & AI. --- Learn more about our lab's leaders in innovation like Karim Lakhani, Alberto Cavallo, Alexander MacKay, Iavor Bojinov, Edward McFowland III, Frank Nagle, and Jacqueline Ng Lane through this video overview presented by Kara Miller. 🎥 #InnovationHub #AIResearch #Innovation #DataScience #AI #ResearchHub

Congratulations Karim! Your leadership continues to elevate the importance of digital upskilling and the potential to generate solutions to important issues facing business. Society. Digital upskilling and maturity are more likely to gain a competitive advantage and ace several performance metrics, like revenue growth, time to market, cost efficiency, product quality, and customer satisfaction. Leading to society, higher education, virtual universities, and much more.

Learn more about NASA's collaboration with IBM to develop INDUS, a LLM for scientific research and how we are exploring the use of INDUS in the Science Discovery Engine. https://lnkd.in/eaUY9x62

"Working together, NASA and IBM Research have developed a new artificial intelligence model to support a variety of weather and climate applications. The new model – known as the Prithvi-weather-climate foundational model – uses artificial intelligence (AI) in ways that could vastly improve the resolution we’ll be able to get, opening the door to better regional and local weather and climate models.  Foundational models are large-scale, base models which are trained on large, unlabeled datasets and can be fine-tuned for a variety of applications. The Prithvi-weather-climate model is trained on a broad set of data – in this case NASA data from NASA’s Modern-Era Retrospective analysis for Research and Applications (MERRA-2)– and then makes use of AI learning abilities to apply patterns gleaned from the initial data across a broad range of additional scenarios." https://lnkd.in/dxNZ7fsE

Scientists from SETI and NASA have proposed using an LLM to allow extraterrestrials to interact with and explore human knowledge. Current data transmission speeds mean it could take centuries to send large models, but data compression could reduce this to about 20 years. News - Franck Marchis from the SETI Institute and NASA's Ignacio Lopez-Francos recently wrote an op-ed for Scientific American. In it, they talked about using AI to communicate with intelligent extraterrestrial life. Here’s what they’re proposing: instead of a talking robot or sentient computer, they suggested something simpler—putting an interactive historical record on a disk for any intelligent life that finds it. 🚀Where did the idea come from? Back in 1977, when the U.S. launched the Voyager 1 spacecraft, scientists included a 12-inch, gold-plated disc with pictures, music, sounds, and texts about our planet. The then President Jimmy Carter even added a message for any aliens who might With today's advanced technology, Marchis and Lopez-Francos think it's time for an update. Instead of just sending images, songs, and writings, they suggest using an LLM so aliens can ask their own questions and explore our knowledge more deeply. 👽What they envision - According to the scientists, using an LLM could let extraterrestrial beings interact with us and learn about humanity without the delays of space travel. However, they also highlight risks, such as the possibility of hostile beings misusing the information. Plus, we’d need to figure out how to send an AI system that operates without the internet. One option is to follow Voyager 1's example, putting the technology on a drive and sending it out. But with Voyager 1 taking nearly 50 years to travel just 15 billion miles, it wouldn't reach our closest star, Alpha Centauri, until around 3084. The other option is to send a signal with the data needed to transmit an LLM, though this would require transmitting data over vast distances at high speed. 🤔Can this work for real? NASA has installations near the Moon that can transmit data at about 100 megabytes per second, which is comparable to what we can achieve on Earth. However, the scientists say that with current technology, interstellar communication would slow to just 100 bits per second. This means sending a model like Meta’s Llama-3-70B to the next galaxy could take centuries. But with some data-compression techniques, they think this time could be reduced to about 20 years.

Artificial intelligence, used by NASA and neurosurgeons, could remake education - Fox News: Artificial intelligence, used by NASA and neurosurgeons, could remake education  Fox News http://dlvr.it/T8CtCn #ai #artificialintelligence