After 3 years, FINALLY out 🥳 Meta-Designing Quantum Experiments with Language Models by @soerenarlt, Haonan Duan, Felix Li, @sangmichaelxie & @Yuhu_ai_ [paper] arxiv.org/abs/2406.02470 A transformer writes Python to solve classes of quantum design questions simultaneously. Method: Trained on synthetic data of pairs of (quantum state sequence, python code for their generation), the model learns automatically to build generalizations. Exciting Results: - Correct generalisations for two classes of entanglement structures, for which previously no generalisation was known! - Unexpected generalisations for another set of six quantum systems, that fit our expectations for small cases, but not for large ones. Interesting new physical entanglement structures? Never investigated by humans Final Thoughts: Both the machine and the human researchers can read and write Python. It's an ideal way to communicate complex scientific ideas, we were basically immediately able to understand the underlying tricks and ideas used by the machine. I bet this will be applicable to many other questions in physics too!

After showing a few examples, large language models can translate natural language mathematical statements into formal specifications. We autoformalize 4K theorems as new data to train our neural theorem prover, achieving SOTA on miniF2F! 1/ Paper: arxiv.org/abs/2205.12615