arXiv Science⌕ Search

arXiv subjects

Eric Zou

Publications and source records attributed to Eric Zou.

4 recordsLinked to original sources

Can an LLM dream? Testing the overfitted brain hypothesis on an agent with text memory

Hoel (2021) proposed that dreams distort replayed experience so that the brain does not overfit to recent events. We test this hypothesis on an LLM agent (Claude Opus 5.5) that stores what it learns in a text memory and rewrites that memory each night from the day's conversation transcript. The agent goes through ten simulated days of conversation built from LongMemEval. We corrupt 5, 10, 20 or 40 percent of the transcript's turns before each nightly rewrite in one of two ways. Detail corruption changes one or two details in a turn to plausible wrong ones, or replaces the turn with one by the same speaker from an earlier day. Random-text corruption replaces half of the selected turns with a passage of Moby-Dick and inserts Moby-Dick words into the other half. A size limit on the memory, rising from 19,500 to 117,000 tokens over ten nights, set from human free-recall data, forces it to forget. We learned that corruption did not help the agent apply what it learned in earlier conversations to new tasks in any of 30 runs (ten conditions, three different conversation histories). Replacing details in 40 percent of turns lowered recall from 76 to 60 percent. Random text had no consistent effect on recall, with a paired difference of 0 $\pm$ 2 points at 40 percent, because the model recognized it as corrupted and left it out of memory. Replaced details were stored as fact even when another turn held the correct value, and the agent answered 40 percent of those questions correctly, compared with 75 percent without corruption. These results suggest that a pretrained model can separate corrupted text from real events, so a fair test needs a learner that cannot make this separation as easily, such as a model fine-tuned on the corrupted days or an agent whose days take place in an invented world.

cs.CL↗

Trade, Trees, and Lives

This paper shows a cascading mechanism through which international trade-induced deforestation results in a decline of health outcomes in cities distant from where trade activities occur. We examine Brazil, which has ramped up agricultural export over the last two decades to meet rising global demand. Using a shift-share research design, we first show that export shocks cause substantial local agricultural expansion and a virtual one-for-one decline in forest cover. We then construct a dynamic area-of-effect model that predicts where atmospheric changes should be felt - due to loss of forests that would otherwise serve to filter out and absorb air pollutants as they travel - downwind of the deforestation areas. Leveraging quasi-random variation in these atmospheric connections, we establish a causal link between deforestation upstream and subsequent rises in air pollution and premature deaths downstream, with the mortality effects predominantly driven by cardiovascular and respiratory causes. Our estimates reveal a large telecoupled health externality of trade deforestation: over 700,000 premature deaths in Brazil over the past two decades. This equates to $0.18 loss in statistical life value per $1 agricultural exports over the study period.

econ.GN↗

Marine Microplastics and Infant Health

A century of plastic usage has led to an accumulation of plastic waste in waterways and oceans. Over time, these wastes break down into particles smaller than 5 microns -- or ''microplastics'' -- which can infiltrate human biological systems. Despite decades of research into this emerging source of environmental pollution, there is a paucity of direct evidence on the health impacts of microplastics exposure at a population scale. This paper reports the first empirical link between in-utero microplastic exposure and adverse birth outcomes. Our analysis is based on a compiled dataset of 3 million births that occurred in coastal areas of 15 countries spanning four continents, which we merge with a novel remote-sensing measurements of marine microplastic concentrations. We show that in-utero exposure to microplastics, particularly during the third trimester of pregnancy, leads to a significant increase in the likelihood of low birth weight. A doubling of exposure increases low birth weight hazard by 0.37 per 1,000 births, which implies over 205,000 cases per year globally can be attributed to microplastic exposure. We further show that aerosolization -- whereby microplastic particles become airborne and inhalable due to seawater evaporation -- is an important pathway for health impact, a challenge that is likely to escalate as ocean temperatures continue to rise.

econ.GN↗

Learning a compass spin model with neural network quantum states

Neural network quantum states provide a novel representation of the many-body states of interacting quantum systems and open up a promising route to solve frustrated quantum spin models that evade other numerical approaches. Yet its capacity to describe complex magnetic orders with large unit cells has not been demonstrated, and its performance in a rugged energy landscape has been questioned. Here we apply restricted Boltzmann machines and stochastic gradient descent to seek the ground states of a compass spin model on the honeycomb lattice, which unifies the Kitaev model, Ising model and the quantum 120$^\circ$ model with a single tuning parameter. We report calculation results on the variational energy, order parameters and correlation functions. The phase diagram obtained is in good agreement with the predictions of tensor network ansatz, demonstrating the capacity of restricted Boltzmann machines in learning the ground states of frustrated quantum spin Hamiltonians. The limitations of the calculation are discussed. A few strategies are outlined to address some of the challenges in machine learning frustrated quantum magnets.

cond-mat.str-el↗