AIThe argument that large language models cannot invent anything genuinely new
A post says Oxford researchers Teppo Felin and Matthias Holweg contrast LLMs’ reliance on existing data with humans’ ability to test theories and create new evidence.
TLDR
A post describing Teppo Felin and Matthias Holweg’s paper “Theory Is All You Need” says the Oxford researchers argue that LLMs cannot invent anything genuinely new because they predict from existing data. The post says the paper uses the Wright brothers as its central example of humans developing a theory and conducting experiments to generate evidence that did not previously exist. It presents LLMs as powerful tools for remixing and accelerating existing knowledge, rather than independent inventors.
Combined views
17.8K
1 Source, first seen 1d ago
The argument that large language models cannot invent anything genuinely new
A post says Oxford researchers Teppo Felin and Matthias Holweg contrast LLMs’ reliance on existing data with humans’ ability to test theories and create new evidence.