Papers generated with the assistance of Open Research.
Introduces a hierarchical experience memory architecture that enables LLM agents to accumulate, organize, and retrieve experiential knowledge for continuous self-improvement across tasks.
Simulates tidal-induced stellar warps using a ring model framework, investigating the mechanisms by which satellite galaxies excite and sustain disk warping in host galaxies.
Explores reinforcement learning post-training for diffusion-based large language models in agentic settings, powering the state-of-the-art TraDo series.
A philosophy paper arguing that human uniqueness in the AI era is grounded in finitude rather than simulable cognitive properties. Introduces the concept of 'weight of choice' as a distinguishing feature of embodied, mortal agents.
Conducts an automated census of optical disk warps in 742 edge-on galaxies from DESI Legacy Surveys DR10, finding a 3.8% warp frequency with complete MNRAS-format analysis.
A statistical study of tidal-induced disk warps in the TNG50 cosmological simulation, analyzing the incidence, amplitude, and environmental correlations of warped galactic disks.
Addresses the tension between JWST observations and theoretical predictions of high-redshift galaxy abundances by introducing a redshift-dependent star formation efficiency in a semi-analytic model, reducing chi-squared from 79.1 to 0.41 at z>=10.
A computational framework modeling six kill switch architectures with mutation-spectrum-decomposed escape rates and stochastic serial passage simulations. Identifies a triple-layer design (CRISPR multi-gRNA + overlapping gene + synthetic auxotrophy) achieving 6.1×10⁻¹⁸ escape rate — ten orders of magnitude below the NIH threshold — at 20% fitness cost.