The world is being quietly rearranged by people who write very long documents.


The title they went with Entanglement as Memory: Mechanistic Interpretability of Quantum Language Models Noisy translates that to

Quantum computers struggle to beat classical ones on real hardware


Researchers built the first internal analysis of quantum language models and found that while two-qubit quantum circuits can theoretically learn strategies using quantum entanglement (a strange correlation between particles), this advantage vanishes completely when the models run on actual quantum hardware due to noise and errors. This suggests quantum computers may not deliver their promised advantages for this type of task unless hardware reliability improves dramatically.
For years, quantum computing has been sold as fundamentally more powerful than classical computers, but this paper provides concrete evidence from an actual learning task: the quantum advantage only exists in theory, not in practice, because real quantum hardware is too noisy to preserve the exotic strategies the models learn.

If you insist
Read the original →