Astrophysics Breakthrough: LLMs as Expert Astronomers A study co-led by the University of Oxford and Google Cloud Partners demonstrated how a general-purpose Large Language Model (LLM) specifically #Gemini can be transformed into an expert astronomy assistant with minimal training. Key Achievement: Using just 15 example images and a simple set of instructions, the multimodal AI learned to accurately classify real changes in the night sky (like exploding stars or black holes tearing stars apart) with approximately 93% accuracy. Crucial Feature: The AI also provided a plain-English explanation for every classification, an important step toward making AI-driven science more transparent and trustworthy. This democratizes scientific discovery, allowing a broader range of scientists to create their own classifiers without deep expertise in training neural networks.

Question 01 AI DeepSeek: AI How does quantum mechanics approach the 11 dimensions of string theory? String Theory: Proposes that fundamental entities are not point particles, but one-dimensional strings vibrating in 11 dimensions (10 spatial + 1 temporal). Quantum Mechanics: Describes the behavior of matter and energy at subatomic scales. How QM Addresses the 11 Dimensions 1. Quantization of Strings Quantum mechanics is applied to the strings themselves—their vibrations are quantized, with different vibrational modes corresponding to different particles: E = ℏω × (vibrational quantum number) 2. Compactified Extra Dimensions Of the 11 dimensions, only 4 are "visible" (3 spatial + time). The other 7 are compactified—curled up to Planck-scale dimensions (~10⁻³⁵ m). 3. Wheeler-deWitt Equation In the quantum approach, a Schrödinger-type equation is applied to the entire 11-dimensional configuration space: ĤΨ[gμν, fields] = 0 Where Ψ is the wave function of the Universe in 11 dimensions. 4. Conceptual Problems · Interpretation: What does a wave function mean for the entire universe? · Measurement: Without an "external observer," how does the wave function collapse? · Background Dependency: Many formulations assume a fixed background, contrary to the spirit of general relativity. Recent Developments · M-Theory: Unification of the 5 versions of string theory via 11 dimensions · AdS/CFT: Correspondence that relates 11D gravity to 10D quantum field theory · Loop Quantum Gravity: Alternative approach that also predicts emergent extra dimensions The creative tension between these theories continues to yield profound insights into the nature of reality! Continued in question 02 AI DeepSeek … #MohammedbinSalman #KSA 🇸🇦 #saudivision2030🇸🇦 #TheRedSea 🇸🇦 #Neom 🇸🇦#redseaproject 🇸🇦 #saudiarabia 🇸🇦 #NASA 🇺🇸 #World 🌍 #Planet 🗺️ #G20 🌍 #UN 🗺️ #G20saudiarabia 🇸🇦 #United Arab Emirates 🇦🇪 #USA 🇺🇲 #Tunisia 🇹🇳 #Palestine 🇵🇸 #Israel 🇮🇱 #Iraq 🇮🇶 #Brazil 🇧🇷 #Argentina 🇦🇷 #Chile 🇨🇱 #Colombia 🇨🇴 #Paraguay 🇵🇾 #Qatar 🇶🇦 #Yemen 🇾🇪 #Oman 🇴🇲 #Kuwait 🇰🇼 #Jordan 🇯🇴 #Bahrain 🇧🇭 #UnitedKingdom 🇬🇧 #France 🇫🇷 #Italy 🇮🇹 #Germany 🇩🇪 #China 🇨🇳 #Russia 🇷🇺 #India 🇳🇪 #bangladesh 🇧🇩 #Japan 🇯🇵 #expo2020dubai #embassy #economy #technology #innovation #AI #AI #design #leadership #growth #education #greenteam ♻️ #theworld #greenpeacebrazil #savetheamazon 🇧🇷 #climate www.greenpeace.org.br #stayathome #wearamask #earthday #digitalhome #digital #savetheworld #blacklivesmatter www.embracyt.com.br

A new study just dropped benchmarking large language models on the International Olympiad on Astronomy & Astrophysics (IOAA) exams — and the results are remarkable. GPT-5 and Gemini 2.5 Pro achieved gold medal–level performance, averaging 85–84%, ranking among the top human competitors out of ~200–300 students worldwide. What’s even more interesting is that while models are acing structured theory exams (multi-step derivations, conceptual reasoning), they still stumble more on open-ended data analysis problems — where real-world messiness and reasoning under uncertainty come into play. As someone working at the intersection of Data Science, NLP, and AI, this resonates deeply. It’s not just about models being “smart” — it’s about identifying the gaps: conceptual reasoning under pressure interpreting messy, real-world signals balancing precision with intuition That’s where the frontier really is. And it’s a reminder: we’re getting closer, but true autonomous research assistants in science aren’t here just yet. Still — the fact that AI can now sit the same exams as top high-school astronomers and medal? That’s a big moment. 🌌 Curious what you think: Do you see LLMs as collaborators in science soon, or are we still in “very smart calculator” territory?

🚀 AI Meets the Cosmos: A New Milestone in Scientific Discovery This week marks a fascinating step forward where artificial intelligence and astrophysics truly converge. Researchers from Oxford University, Google Cloud, and Radboud University have unveiled a breakthrough: a general-purpose AI model (Gemini) capable of detecting and explaining cosmic phenomena such as supernovae, black hole flares, and fast-moving asteroids, using only a handful of examples. Why this matters: Few-shot learning at scale: The model achieved about 93% accuracy after training on roughly 15 images, a remarkable leap toward data efficiency. Explainability built in: It doesn’t just classify cosmic events; it explains its reasoning in clear, human-understandable language. Democratizing research: Scientists without massive AI infrastructure can now develop their own models to detect astronomical events. Beyond astronomy: The same techniques could transform other domains, from medical imaging and climate modeling to anomaly detection in cybersecurity. This is a strong reminder that innovation rarely follows a straight path. The real breakthroughs often emerge when distinct disciplines like AI and astrophysics intersect. As someone deeply interested in the synergy between AI, sensing, and interpretability, I find this development incredibly inspiring. The future of intelligent scientific discovery isn’t just near, it’s unfolding right now.

Inside MIT’s New AI Platform for Scientific Discovery. https://lnkd.in/gjudBVVb. Lead: Cosmologists can now explore data faster than ever before with a new emulator. https://lnkd.in/gSwtkqpr

Ever thought about how artificial intelligence and physics are teaming up to explore the universe? It still blows my mind that AI can now help spacecraft make decisions in real time — choosing routes, analyzing terrain, and predicting system failures millions of miles away where human help just can’t reach fast enough. In simple terms: physics builds the rocket, AI teaches it how to think. Reinforcement learning models (basically trial-and-error learning for machines) are already being used to train rovers to move smarter and safer. It’s like giving curiosity a brain — and a bit of courage. As someone studying Applied Physics and Computer Science at ASU, I’ve started realizing how these two fields aren’t separate anymore. The future of exploration — on Mars or beyond — might depend on how well we get them to talk to each other. And honestly… that’s the kind of future I want to help build. What do you think — could AI ever replace human intuition in discovery, or will it just make our curiosity limitless? #AI #Physics #SpaceExploration #MachineLearning #Innovation #STEM #Curiosity #adiscoveryinsideimmortalized #Philaquest

✨ A new AI tool developed by Google Cloud and University of Oxford researchers has been able to accurately classify astronomical events and explain its reasoning. 🔭 The work is being hailed as a significant breakthrough due to the tiny amount of data needed to train the AI tool. The researchers used just 15 astronomical images and a simple set of instructions, prompting Google’s Gemini large language model (LLM) to learn to distinguish real cosmic events from false positives. 🌜 Following these minimal instructions, Gemini was able to detect real cosmic events with over 93 per cent accuracy, while also providing a clear and concise explanation for each classification. Read more here 👇 https://lnkd.in/da7WvMTT #TheEngineerUK #ArtificialIntelligence #AIResearch #MachineLearning #SpaceScience #Astronomy #DeepLearning #BigData #GoogleGemini #Innovation #ScientificDiscovery #Astrophysics #DataScience #LLM #TechBreakthrough #AIinScience

This is a great post from investor in Periodic Labs. Esp for me to read as a former physical chemist. Kudos Slater Stich and Bain Capital Ventures (BCV). Also aligns with our thesis that AI is asymptotically reducing the cost of intelligence. “Typically, only the successful experiments are published — for every synthesis we know about, there are probably 10x or 100x more failed attempts that never saw their day in a journal. There are only so many grad students, meaning only so many experiments. Going back to Ulam, you start thinking differently if you can easily do 4,000,000 multiplications instead of painstakingly doing 4,000 by slide rule. Similarly, you might start thinking differently if you could run radically more solid state syntheses.” https://lnkd.in/eCP_JyU9

Thrilled to share that our research paper titled “Exploring Distance Metrics for Early RUL Estimation: a Liquid Neural Network Approach” is now published on IEEE Xplore! Presented at ACTEA 2025, this study focuses on improving the accuracy and interpretability of Remaining Useful Life (RUL) prediction models for turbofan engines — a key challenge in predictive maintenance. 🧠 What we tackled: We examined how different distance metrics — Cosine, Manhattan, Euclidean, and Chebyshev — influence the initial RUL threshold, a factor often overlooked yet critical to model reliability. ⚙️ Our approach: A Liquid Neural Network (LNN) trained on NASA’s C-MAPSS dataset was used to model engine degradation through adaptive, continuous-time learning. This research was authored by Haidar Damen and co-authored by Dr. Michel Owayjan, Gaby Abou Haidar - Ph.D., Roger R. Achkar, Ph.D, whose combined efforts made this publication possible. 📄 Read the full paper on IEEE Xplore: https://lnkd.in/eXtwZNNz #MachineLearning #LiquidNeuralNetworks #PredictiveMaintenance #RUL #AI #IEEE #ACTEA2025