Machina Mirabilis: Can an LLM Discover Quantum Mechanics and Relativity?
An experiment to see if a large language model (LLM) trained solely on pre-1900 text can independently discover principles of quantum mechanics and relativity.
Why it matters
This experiment explores the limits of what large language models can achieve in terms of scientific discovery and reasoning.
Key Points
- 1Researchers trained an LLM from scratch on historical texts prior to 1900
- 2The goal was to see if the LLM could rediscover modern physics concepts
- 3The experiment aimed to explore the potential of LLMs for open-ended discovery
Details
The Machina Mirabilis experiment explores whether a large language model (LLM) trained exclusively on historical texts from before 1900 can independently discover the fundamental principles of quantum mechanics and relativity. The researchers hypothesized that by training the LLM on a vast corpus of pre-20th century knowledge, it may be able to piece together the underlying laws of physics through patterns in the text, without any explicit training on modern scientific theories. This experiment pushes the boundaries of what LLMs are capable of in terms of open-ended discovery and scientific reasoning, rather than just task-specific applications. The results could shed light on the potential of LLMs to serve as tools for scientific exploration and the generation of novel ideas.
No comments yet
Be the first to comment