Human-LLM Coevolution: Evidence from Academic Writing

With a statistical analysis of arXiv paper abstracts, we report a marked drop in the frequency of several words previously identified as overused by ChatGPT, such as "delve", starting soon after they were pointed out in early 2024. The frequency of certain other words favored by ChatGPT, such as "significant", has instead kept increasing. These phenomena suggest that some authors of academic papers have adapted their use of large language models (LLMs), for example, by selecting outputs or applying modifications to the LLM-generated content. Such coevolution and cooperation of humans and LLMs thus introduce additional challenges to the detection of machine-generated text in real-world scenarios. Estimating the impact of LLMs on academic writing by examining word frequency remains feasible, and more attention should be paid to words that were already frequently employed, including those that have decreased in frequency.

Are Large Language Models Chameleons?

Do large language models (LLMs) have their own worldviews and personality tendencies? Simulations in which an LLM was asked to answer subjective questions were conducted more than 1 million times. Comparison of the responses from different LLMs with real data from the European Social Survey (ESS) suggests that the effect of prompts on bias and variability is fundamental, highlighting major cultural, age, and gender biases. Methods for measuring the difference between LLMs and survey data are discussed, such as calculating weighted means and a new proposed measure inspired by Jaccard similarity. We conclude that it is important to analyze the robustness and variability of prompts before using LLMs to model individual decisions or collective behavior, as their imitation abilities are approximate at best.

floZ: Evidence estimation from posterior samples with normalizing flows

We propose a novel method (floZ), based on normalizing flows, for estimating the Bayesian evidence (and its numerical uncertainty) from a set of samples drawn from the unnormalized posterior distribution. We validate it on distributions whose evidence is known analytically, up to 15 parameter space dimensions, and compare with two state-of-the-art techniques for estimating the evidence: nested sampling (which computes the evidence as its main target) and a k-nearest-neighbors technique that produces evidence estimates from posterior samples. Provided representative samples from the target posterior are available, our method is more robust to posterior distributions with sharp features, especially in higher dimensions. It has wide applicability, e.g., to estimate the evidence from variational inference, Markov-chain Monte Carlo samples, or any other method that delivers samples from the unnormalized posterior density.

Is ChatGPT Transforming Academics' Writing Style?

Based on one million arXiv papers submitted from May 2018 to January 2024, we assess the textual density of ChatGPT's writing style in their abstracts by means of a statistical analysis of word frequency changes. Our model is calibrated and validated on a mixture of real abstracts and ChatGPT-modified abstracts (simulated data) after a careful noise analysis. We find that ChatGPT is having an increasing impact on arXiv abstracts, especially in the field of computer science, where the fraction of ChatGPT-revised abstracts is estimated to be approximately 35%, if we take the output of one of the simplest prompts, "revise the following sentences", as a baseline. We conclude with an analysis of both positive and negative aspects of the penetration of ChatGPT into academics' writing style.

Stratified Learning: a general-purpose statistical method for improved learning under Covariate Shift

Covariate shift arises when the labelled training (source) data is not representative of the unlabelled (target) data due to systematic differences in the covariate distributions. A supervised model trained on the source data subject to covariate shift may suffer from poor generalization on the target data. We propose a novel, statistically principled and theoretically justified method to improve learning under covariate shift conditions, based on propensity score stratification, a well-established methodology in causal inference. We show that the effects of covariate shift can be reduced or altogether eliminated by conditioning on propensity scores. In practice, this is achieved by fitting learners on subgroups ("strata") constructed by partitioning the data based on the estimated propensity scores, leading to balanced covariates and much-improved target prediction. We demonstrate the effectiveness of our general-purpose method on contemporary research questions in observational cosmology, and on additional benchmark examples, matching or outperforming state-of-the-art importance weighting methods, widely studied in the covariate shift literature. We obtain the best reported AUC (0.958) on the updated "Supernovae photometric classification challenge" and improve upon existing conditional density estimation of galaxy redshift from Sloan Data Sky Survey (SDSS) data.