Shi Feng

Apr 17, 2024 | lesswrong.com | Arjun Panickssery |Sam Bowman |Shi Feng

Self-evaluation using LLMs is used in reward modeling, model-based benchmarks like GPTScore and AlpacaEval, self-refinement, and constitutional AI. LLMs have been shown to be accurate at approximating human annotators on some tasks. But these methods are threatened by self-preference, a bias in which an LLM evaluator scores its own outputs higher than than texts written by other LLMs or humans, relative to the judgments of human annotators.

Apr 15, 2024 | arxiv.org | Samuel R. Bowman |Shi Feng

Mar 18, 2024 | link.aps.org | Shi Feng |Adhip Agarwala |Nandini Trivedi

We investigate the fate of the Kitaev spin liquid (KSL) under the influence of an external magnetic field h in the [001] direction and upon tuning bond anisotropy of the Kitaev coupling Kz keeping Kx=Ky=K. Guided by density matrix renormalization group, exact diagonalization, and with insights from parton mean-field theory, we uncover a field-induced gapless-to-gapless Lifshitz transition from the nodal KSL to an intermediate gapless phase.

Feb 23, 2024 | link.aps.org | Shi Feng

We reveal the connection between two-dimensional subsystem symmetry-protected topological (SSPT) states and two-dimensional topological orders via a self-dual frustrated toric code model. This model, an enrichment of the toric code (TC) with its dual interactions, can be mapped to a model defined on the dual lattice with subsystem symmetries and subextensive ground state degeneracy.

Aug 28, 2023 | chine.in | Li Yue |Shi Feng

Un homme ramasse des déchets dans la rivière Songhua, à Harbin, dans la province chinoise du Heilongjiang (nord-est), le 27 août 2023.