賀沢 秀人氏 (Ph.D., Google ソフトウェアエンジニア / 言語処理学会 理事)

[KEK連携コロキウム] 大規模言語モデルの舞台裏

##### 最近 ChatGPT に代表される大規模言語モデル (Large Language Model; LLM）が大きな話題となっている。LLM以前の自動対話システムでは、宅配便の配達問い合わせのようにあらかじめ決められた話題と目的のために応答を生成するのがやっとであり、しかもその応答も定型文の域を出ないものが多かった。それにたいしLLMは一見人間が応答しているのと区別がつかないやりとりを広範な話題について行うことができるため、様々な応用が期待されるとともに、その発展についていろいろな懸念も生じている。本講演では、LLMとは実際のところどういうものなのかその仕組みを解説するとともに、単なる印象論を越えてLLMについて冷静に議論をするための視点を提供する。

Diego Blas, UAB

Detecting (high frequency) gravitational waves in a box

##### In the presence of a background EM field, the passage of GWs generate different modes in EM cavities by either their effect on the background field or by interacting with the boundaries of the cavity. In this talk, I will describe this process in some detail and show that these set-ups provide some of the best bounds for GWs in the MHz-GHz band. I shall finish by briefly describing the way GWs interact with spin systems.

Yusuke Taki, Kyoto University

Entropy in dS/CFT correspondence and its application

##### In this talk, we focus on dS/CFT correspondence, which is a proposal for de Sitter (dS) holography analogous to AdS/CFT. This talk contains two recent works. First we consider an application of dS/CFT to evaluation of the semi-classical limit of wave function of universe. Through the procedure of analytic continuation in the CFT side corresponding to that from AdS to dS, we can find an interesting structure that only “allowable” geometries appear as saddle points of the gravitational path integral. Second, we discuss an extension of the holographic entanglement entropy formula in AdS/CFT toward dS/CFT. We propose that the holographic formula is formulated for a generalized notion of entanglement entropy, called pseudo entropy.

Maki Takeuchi, Kobe University

The Mystery of the Standard Model and the Extra-Dimensional Model

##### The Standard Model achieved significant success with the discovery of the Higgs in 2012. However, there are still numerous unexplained phenomena. One of them is the number of generations problem, where quarks and leptons come in three copies with only differences in mass while having identical spin and charge. Is the existence of these three generations a mere coincidence, or is there a deeper reason behind it? The Standard Model cannot explain why there are three generations.On the other hand, the higher dimensional theory offers a possible explanation for the number of generations problem. In the higher dimensional theory, the number of generations is related to the geometry of the extra dimensions. In other words, the number of generations carries a physical meaning associated with the geometry of the extra dimensions. In this seminar, I will talk about the generation structure of the T^2/Z_N orbifold with magnetic flux.

Joe Sato, Yokohama National U

L_{\mu-\tau} symmetry and its unification with Standard gauge group

##### In this talk I show many aspects of gauged L_{\mu-\tau} symmetry. The symmetry gauges {the muon number – the tau number} in the standard model so as to be anomaly free. First I will show how it is interesting for phenomenology. It includes IceCube Gap, Hubble tension etc. Then I show our phenomenological model for lepton mass and mixing with the symmetry. In this model the symmetry appears as the family symmetry. Finally I will

present my trial to unify the symmetry with the Standard gauge group in terms of coset space family unification.

Kazumi Okuyama, Shinshu University

Hartle-Hawking wavefunction in double scaled SYK

##### We compute the transition amplitude between the chord number 0 and ￥ ell states in the double scaled SYK model and interpret it as a Hartle-Hawking wavefunction of the bulk gravitational theory. We observe that the so-called un-crossed matter correlators of double scaled SYK model are obtained by gluing the Hartle-Hawking wavefunctions with an appropriate weight.

Kotaro Murakami, Tokyo Institute of Technology

Investigation of baryon resonances from meson-baryon scatterings in lattice QCD

##### Studying hadron resonances in lattice QCD plays an important role in understanding exotic hadrons. We investigate baryon resonances from meson-baryon scatterings in the HAL QCD method, where we derive scattering amplitudes in lattice QCD via the interaction potentials. In this talk, I present the analysis of Δ and Ω baryons, both of which belong to the baryon decuplet. In our analysis, we use heavy quark masses so that Δ is a stable particle as well as Ω. If time permits, I also show our preliminary results of the study on Λ(1405) in flavor SU(3) limit.

Alexander Broll, Humboldt University of Berlin

Estimating the B Pi Excited States Contamination of B Meson Correlators with Heavy Meson Chiral Perturbation Theory

##### CKM matrix elements are free parameters of the Standard Model which are determined by combining experimental and theoretical input. On the theory side, this requires the computation of hadronic matrix elements (ME) in lattice simulations. The lattice correlators can suffer from excited states contamination, i.e. additional contributions from multi-particle states with the same quantum numbers as the initial and/or final state of the matrix element. This can lead to an over- or underestimation of the ME and thus to a systematic error in the results of CKM matrix elements. In this talk, I present how Heavy Meson Chiral Perturbation Theory can be used to estimate the excited states contamination of B meson correlation functions relevant for flavour physics.

Masamichi Miyaji, Nagoya University

Fluctuations in the Entropy of Hawking Radiation

##### Recent study revealed that the inclusion of Euclidean wormhole into the gravitational path integral renders the entropy of Hawking radiation consistent with unitarity, deriving the Page curve of the Hawking radiation. On the other hand, since the gravitational path integral with Euclidean wormhole computes quantities of ensemble average of theories, it is possible that the entropy of Hawking radiation of each gravity theory fluctuate wildly around the ensemble average. In this talk we show that such fluctuation is as small as the dimension of the system, ensuring the answer from the ensemble average is typical. We use the gravitational path integral to compute the fluctuations of the Hawking radiation entropy around the Page curve, in a two-dimensional model introduced by Penington \emph{et al}. Before the Page time, we find that $\delta S = e^{-S}/\sqrt{2}$, where $S$ is the black hole entropy. This result agrees with the Haar-averaged entropy fluctuations of a bipartite system, which we also compute at leading order. After the Page time, we find that $\delta S = \sqrt{2}e^{-S}/\pi$. This is not symmetric under exchange of subsystem sizes and so does not agree with the Haar average for a subsystem of fixed Hilbert space dimension. We show that the discrepancy can be attributed to an additive $\sqrt{2}/\pi$ fluctuation in the number of black hole states in a given energy band. As a by- product, our result gives a refinement on the known upper bound on the subsystem entropy fluctuation in Haar random pure state.

Owe Philipsen, Frankfurt University

[QCD theory Seminar] Surprises on the way toward the QCD phase diagram