arXiv subjects

Kirill Kalinin

Publications and source records attributed to Kirill Kalinin.

7 recordsLinked to original sources

arXiv2026

Reflections and New Directions for Human-Centered Large Language Models

Large Language Models (LLMs) are increasingly shaping the private and professional lives of users, with numerous applications in business, education, finance, healthcare, law, and science. With this rise in global influence comes greater urgency to build, evaluate, and deploy these systems in a manner that prioritizes not only technical capabilities but also human priorities. This work presents a framework for developing Human-Centered Large Language Models (HCLLMs), which integrates perspectives from Natural Language Processing (NLP), Human-Computer Interaction (HCI), and responsible AI. Considering the ethics, economics, and technical objectives of language modeling, we argue that model developers need to address human concerns, preferences, values, and goals, not only during a cursory post-training stage, but rather with rigor and care at every stage of the pipeline. This paper offers human-centered insights and recommendations for developers at each stage, from system design to data sourcing, model training, evaluation, and responsible deployment. Then we conclude with a case study, applying these insights to understand the future of work with HCLLMs.

cs.CL↗

arXiv2026

Learning State-Tracking from Code Using Linear RNNs

Over the last years, state-tracking tasks, particularly permutation composition, have become a testbed to understand the limits of sequence models architectures like Transformers and RNNs (linear and non-linear). However, these are often sequence-to-sequence tasks: learning to map actions (permutations) to states, which is incompatible with the next-token prediction setting commonly used to train language models. We address this gap by converting permutation composition into code via REPL traces that interleave state-reveals through prints and variable transformations. We show that linear RNNs capable of state-tracking excel also in this setting, while Transformers still fail. Motivated by this representation, we investigate why tracking states in code is generally difficult: actions are not always fully observable. We frame this as tracking the state of a probabilistic finite-state automaton with deterministic state reveals and show that linear RNNs can be worse than non-linear RNNs at tracking states in this setup.

cs.LG↗

arXiv2017

Matter wave coupling of spatially separated and unequally pumped polariton condensates

Spatial quantum coherence between two separated driven-dissipative polariton condensates created non-resonantly and with a different occupation is studied. We identify the regions where the condensates remain coherent with the phase difference continuously changing with the pumping imbalance and the regions where each condensate acquires its own chemical potential with phase differences exhibiting time-dependent oscillations. We show that in the mutual coherence limit the coupling consists of two competing contributions: a symmetric Heisenberg exchange and the Dzyloshinskii-Moriya asymmetric interactions that enable a continuous tuning of the phase relation across the dyad and derive analytic expressions for these types of interactions. The introduction of non-equal pumping increases the complexity of the type of the problems that can be solved by polariton condensates arranged in a graph configuration. If equally pumped polaritons condensates arrange their phases to solve the constrained quadratic minimisation problem with a real symmetric matrix, the non-equally pumped condensates solve that problem for a general Hermitian matrix.

cond-mat.mes-hall↗

arXiv2017

Geometric frustration in polygons of polariton condensates creating vortices of varying topological charge

Vorticity is a key ingredient to a broad variety of fluid phenomena, and its quantised version is considered to be the hallmark of superfluidity. Circulating flows that correspond to vortices of a large topological charge, termed giant vortices, are notoriously difficult to realise and even when externally imprinted, they are unstable, breaking into many vortices of a single charge. In spite of many theoretical proposals on the formation and stabilisation of giant vortices in ultra-cold atomic Bose-Einstein condensates and other superfluid systems, their experimental realisation remains elusive. Polariton condensates stand out from other superfluid systems due to their particularly strong interparticle interactions combined with their non-equilibrium nature, and as such provide an alternative testbed for the study of vortices. Here, we non-resonantly excite an odd number of polariton condensates at the vertices of a regular polygon and we observe the formation of a stable discrete vortex state with a large topological charge as a consequence of antibonding frustration between nearest neighbouring condensates.

cond-mat.mes-hall↗

arXiv2017

Exotic states of matter with polariton chains

We consider linear periodic chains of exciton-polariton condensates formed by pumping polaritons non-resonantly into a linear network. To the leading order, such a sequence of condensates establishes relative phases as to minimize a classical one-dimensional XY Hamiltonian with nearest and next to nearest neighbors. We show that the low energy states of polaritonic linear chains demonstrate various classical regimes: ferromagnetic, antiferromagnetic and frustrated spiral phases, where quantum or thermal fluctuations are expected to give rise to a spin liquid state. At the same time nonlinear interactions at higher pumping intensities bring about phase chaos and novel exotic phases.

cond-mat.mes-hall↗

arXiv2017

Simulating the spectral gap with polariton graphs

We recently proposed polariton graphs as a novel platform for solving hard optimization problems that can be mapped into the $XY$ model. Here, we elucidate a relationship between the energy spectrum of the $XY$ Hamiltonian and the total number of condensed polariton particles. Using as a test-bed the hexagonal unit lattice we show that the lower energy states of the $XY$ Hamiltonian are faithfully reproduced by mean-field numerical simulations utilising the Ginzburg--Landau equation coupled to an exciton reservoir. Our study paves the way to simulating the spectral gap of the XY model using polariton graphs.

cond-mat.other↗

arXiv2016

Realizing the $XY$ Hamiltonian in polariton simulators

Several platforms are currently being explored for simulating physical systems whose complexity increases faster than polynomially with the number of particles or degrees of freedom in the system. Defects and vacancies in semiconductors or dielectric materials, magnetic impurities embedded in solid helium \cite{lemeshko13}, atoms in optical lattices, photons, trapped ions and superconducting q-bits are among the candidates for predicting the behaviour of spin glasses, spin-liquids, and classical magnetism among other phenomena with practical technological applications. Here we investigate the potential of polariton graphs as an efficient simulator for finding the global minimum of the $XY$ Hamiltonian. By imprinting polariton condensate lattices of bespoke geometries we show that we can simulate a large variety of systems undergoing the U(1) symmetry breaking transitions. We realise various magnetic phases, such as ferromagnetic, anti-ferromagnetic, and frustrated spin configurations on unit cells of various lattices: square, triangular, linear and a disordered graph. Our results provide a route to study unconventional superfluids, spin-liquids, Berezinskii-Kosterlitz-Thouless phase transition, classical magnetism among the many systems that are described by the $XY$ Hamiltonian.

cond-mat.mes-hall↗