arXiv subjects

Jorge Morales

Publications and source records attributed to Jorge Morales.

10 recordsLinked to original sources

arXiv2025

Artificial Phantasia: Emergent Mental Imagery in Large Language Models

Can visual imagery be driven solely by language? This idea goes against cognitive science's traditional view that visual mental imagery is only possible through pictorial representations. Large Language Models (LLMs) provide nascent evidence not only that visual mental imagery via propositional-representations is possible, but that it can be more robust than human imagination. We created dozens of novel items for an extension to a classic task which is argued to be solvable exclusively via pictorial representations (i.e., language alone would be insufficient). Subjects were asked to imagine a series of compositional letter and shape transformations and identify the resultant "image". We found that the best LLMs performed significantly better than humans ($n = 100$ human participants, $p < .0001$), indicating the existence of an artificial phantasia, or emergent "visual" mental imagery that may not be pictorial. Furthermore, we tested reasoning models with variable reasoning-token allocation and found that models perform best with longer reasoning chains, demonstrating a linguistic impact on the task -- language alone may be sufficient. We examined three emergent imagery hypotheses: pure propositional imagery, propositional imagery with visio-linguistic priors, or pictorial visual imagery (classical visual imagery). Our study not only presents evidence for a previously unreported emergent cognitive capacity of LLMs, but also reignites debate on the requirement for a pictorial format in mental imagery.

cs.AI↗

arXiv2025

Self-Interpretability: LLMs Can Describe Complex Internal Processes that Drive Their Decisions

We have only limited understanding of how and why large language models (LLMs) respond in the ways that they do. Their neural networks have proven challenging to interpret, and we are only beginning to tease out the function of individual neurons and circuits within them. However, another path to understanding these systems is to investigate and develop their capacity to explain their own functioning. Here, we show that i) LLMs can accurately describe quantitative features of their own internal processes during certain kinds of decision-making and ii) that it is possible to improve these capabilities through training. To do so, we fine-tuned GPT-4o and GPT-4o-mini to make decisions in a wide variety of complex contexts (e.g., choosing between condos, loans, vacations, etc.) according to randomly-generated, quantitative preferences about how to weigh different attributes (e.g., the relative importance of natural light versus quiet surroundings for condos). We demonstrate that the LLMs can accurately report these preferences (i.e., the weights that they learned to give to different attributes during decision-making). Next, we demonstrate that these LLMs can be fine-tuned to explain their decision-making even more accurately. Finally, we demonstrate that this training generalizes: It improves the ability of the models to accurately explain how they make other complex decisions, not just decisions they have been fine-tuned to make. This work is a step towards training LLMs to accurately and broadly report on their own internal processes -- a possibility that would yield substantial benefits for interpretability, control, and safety.

cs.CL↗

arXiv2025

Results from an Einstein@Home search for continuous gravitational waves from Cassiopeia A and Vela Jr. using LIGO O2 data

We conduct two searches for continuous, nearly monochromatic gravitational waves originating from the central compact objects in the supernova remnants Cassiopeia A and Vela Jr. using public LIGO data. The search for Cassiopeia A targets signal frequencies between 20 Hz and 400 Hz; the Vela Jr. search between 400 Hz and 1700 Hz, and both investigate the broadest set of waveforms ever considered with highly sensitive deterministic search methods. Above 1500 Hz the Vela Jr. search is the most sensitive carried out thus far, improving on previous results by over 300\%. Above 976 Hz these results improve on existing ones by 50\%. In all we investigate over $10^{18}$ waveforms, leveraging the computational power donated by thousands of Einstein@Home volunteers. We perform a 4-stage follow-up on more than 6 million waveforms. None of the considered waveforms survives the follow-up scrutiny, indicating no significate detection candidate. Our null results constrain the maximum amplitude of continuous signals as a function of signal frequency from the targets. The most stringent 90\% confidence upper limit for Cas A is $h_0^{90 \%}\approx 7.3\times10^{-26}$ near 200 Hz, and for Vela Jr. it is $h_0^{90 \%}\approx 8.9\times10^{-26}$ near 400 Hz. Translated into upper limits on the ellipticity and r-mode amplitude, our results probe physically interesting regions: for example the ellipticity of Vela Jr. is constrained to be smaller than $10^{-7}$ across the frequency band, with a tighter constraint of less than $2\times10^{-8}$ at the highest frequencies.

gr-qc↗

arXiv2019

Energy Loss Due to Defect Formation from $^{206}$Pb Recoils in SuperCDMS Germanium Detectors

The Super Cryogenic Dark Matter Search experiment (SuperCDMS) at the Soudan Underground Laboratory studied energy loss associated with Frenkel defect formation in germanium crystals at mK temperatures using in situ $^{210}$Pb sources. We examine the spectrum of $^{206}$Pb nuclear recoils near its expected 103 keV endpoint energy and determine an energy loss of $\left(6.08\pm0.18\right)$ %, which we attribute to defect formation. From this result and using TRIM simulations, we extract the first experimentally determined average displacement threshold energy of $\left(19.7^{+0.6}_{-0.5}\right)$ eV for germanium. This has implications for the analysis thresholds of future germanium-based dark matter searches.

physics.ins-det↗

arXiv2016

Modelling of spatial structure of divertor footprints caused by edge-localized modes mitigated by magnetic perturbations

Resonant magnetic perturbations (RMPs) can mitigate the edge-localized modes (ELMs), i.e. cause a change of the ELM character towards smaller energy loss and higher frequency. During mitigation a change of the spatial structure of ELM loads on divertor was observed on DIII-D and MAST: the power is deposited predominantly in the footprint structures formed by the magnetic perturbation. In the present contribution we develop a theory explaining this effect, based on the idea that part of the ELM loss is caused by parallel transport in the homoclinic tangle formed by the magnetic perturbation of the ELM. The modified tangle resulting from the combination of the ELM perturbation and the applied RMP has the expected property of bringing open field lines in the same areas as the tangle from the RMP alone. We show that this explanation is consistent with features of the mitigated ELMs on MAST. We in addition validated our theory by an analysis of simulations of mitigated ELMs using the code JOREK. We produced detailed laminar plots of field lines on the divertor in the JOREK runs with an ELM, an applied RMP, and an ELM mitigated by the presence of the RMP. The results for an ELM clearly show a high-n rotating footprint structure appearing during the nonlinear stage of the ELM, which is not present in the early stage of the ELM. The results for a n=2 RMP from the ELM control coils show the expected n=2 footprint structure. The results for the mitigated ELM show a similar structure, modulated by a higher n perturbation of the ELM, consistent with our theory.

physics.plasm-ph↗

arXiv2014

Generic extensions and generic Polynomials for multiplicative groups

Let $\mathcal{A}$ be a finite-dimensional algebra over a finite field $\mathbf{F}_q$ and let $G=\mathcal{A}^\times$ be the multiplicative group of $\mathcal{A}$. In this paper, we construct explicitly a generic Galois $G$-extension $S/R$, where $R$ is a localized polynomial ring over $\mathbf{F}_q$, and an explicit generic polynomial for $G$ in $\dim_{\mathbf{F}_q}(\mathcal{A})$ parameters.

math.AG↗

arXiv2012

Intrinsic rotation of toroidally confined magnetohydrodynamics

The spatiotemporal self-organization of viscoresistive magnetohydrodynamics (MHD) in a toroidal geometry is studied. Curl-free toroidal magnetic and electric fields are imposed. It is observed in our simulations that a flow is generated, which evolves from dominantly poloidal to toroidal when the Lundquist numbers are increased. It is shown that this toroidal organization of the flow is consistent with the tendency of the velocity field to align with the magnetic field. Up-down asymmetry of the geometry causes the generation of a non-zero toroidal angular momentum.

physics.plasm-ph↗

arXiv2011

Siegel's mass formula and averages of Dirichlet L-functions over function fields

Let D be a square-free polynomial in F_q[t], where q is odd, and let G be a genus of definite ternary lattices over F_q[t] of determinant D. In this paper we give self-contained and relatively elementary proofs of Siegel's formulas for the weighted sum of primitive representations numbers over the classes of G and for the mass of G. Our proof of the mass formula shows an interesting relation with certain averages of Dirichlet L-functions.

math.NT↗

arXiv2009

Representations of Definite Binary Quadratic Forms over F_q[t]

In this paper, we prove that a binary definite quadratic form over F_q[t], where q is odd, is completely determined up to equivalence by the polynomials it represents up to degree 3m-2, where m is the degree of its discriminant. We also characterize, when q>13, all the definite binary forms over F_q[t] that have class number one.

math.NT↗

arXiv2009

Isospectral Definite Ternary F_q[t]-Lattices

We prove that the representations numbers of a ternary definite integral quadratic form defined over F_q[t], where F_q is a finite field of odd characteristic, determine its integral equivalence class when q is large enough with respect to its successive minima. Equivalently, such a quadratic form is determined up to integral isometry by its theta series.

math.NT↗