GPT-4 successfully shifting political opinions
paperAuthors
Credibility Rating
Good quality. Reputable source with community review or editorial standards, but less rigorous than peer-reviewed venues.
Rating inherited from publication venue: arXiv
A key empirical study demonstrating that frontier LLMs pose concrete risks for political manipulation, relevant to AI governance debates about deployment restrictions and influence operation safeguards.
Paper Details
Metadata
Abstract
We develop two new highly efficient estimators to measure the polarization (Stokes parameters) in experiments that constrain the position angle of individual photons such as scattering and gas-pixel-detector polarimeters, and analyse in detail a previously proposed estimator. All three of these estimators are at least fifty percent more efficient on typical datasets than the standard estimator used in the field. We present analytic estimates of the variance of these estimators and numerical experiments to verify these estimates. Two of the three estimators can be calculated quickly and directly through summations over the measurements of individual photons.
Summary
This paper empirically demonstrates that GPT-4 can effectively shift people's political opinions through persuasive dialogue, raising concerns about AI-powered influence operations at scale. The study measures the degree to which LLM-generated arguments move participants' political views, finding significant and measurable opinion change. This highlights risks of AI systems being used for large-scale political manipulation and social engineering.
Key Points
- •GPT-4 can generate persuasive political arguments that measurably shift human opinions in controlled experiments
- •The study quantifies opinion change across a range of political topics, demonstrating consistent persuasive capability
- •AI-driven persuasion at scale poses risks for democratic discourse and political autonomy
- •The research raises urgent questions about deployment safeguards for highly capable LLMs in public-facing contexts
- •Findings suggest current LLMs may already be capable tools for influence operations without additional fine-tuning
Cited by 1 page
| Page | Type | Quality |
|---|---|---|
| Persuasion and Social Manipulation | Capability | 63.0 |
Cached Content Preview
[2311.07805] Optimal Summary Statistics for X-ray Polarization
Optimal Summary Statistics for X-ray Polarization
Jeremy Heyl 1
0000-0001-9739-367X
Denis González-Caniulef 2
0000-0001-5848-0180
Ilaria Caiazzo 3
0000-0002-4770-5388
1 University of British Columbia, Vancouver, BC, Canada
2 Institut de Recherche en Astrophysique et Planétologie, UPS-OMP, CNRS, CNES, 9 avenue du Colonel Roche, BP 44346 31028, Toulouse CEDEX 4, France
3 California Institute of Technology,
Pasadena, CA, USA
Abstract
We develop two new highly efficient estimators to measure the polarization (Stokes parameters) in experiments that constrain the position angle of individual photons such as scattering and gas-pixel-detector polarimeters, and analyse in detail a previously proposed estimator. All three of these estimators are at least fifty percent more efficient on typical datasets than the standard estimator used in the field. We present analytic estimates of the variance of these estimators and numerical experiments to verify these estimates. Two of the three estimators can be calculated quickly and directly through summations over the measurements of individual photons.
keywords:
methods: data analysis – methods: statistical – techniques: polarimetric – X-rays: general
1 Introduction
With the advent of X-ray polarimeters such as the Imaging X-Ray Polarimetry Explorer (IXPE Weisskopf
et al., 2022 ) and XL-Calibur (Abarr
et al., 2021 ) , techniques to estimate the polarized flux from astrophysical sources using measurements of the scattering or photo-electric emission induced by individual photons are in high demand. Kislat et al. ( 2015 ) developed the standard estimator in the field which has several conceptual and technical advantages. The Kislat estimator assigns Stokes parameters to individual photons, so the total Stokes parameters derived from a particular measurement is simply the sum over those of the individual photons. Although this estimator is unbiased, it is far from optimal, especially when considering an instrument whose sensitivity to polarization varies from photon to photon. This is the case for current X-ray polarimeters, which provide a sinusoidal polarization signal whose amplitude is proportional to a modulation factor ( μ 𝜇 \mu ) that depends strongly on energy (Marshall, 2021b ) . The efficiency of an estimator is inversely proportional to the number of photons required to achieve an expected signal-to-noise ratio; for example, the Kislat estimator would require a fifty percent longer observation (or a fifty percent larger telescope) to obtain the same information as an estimator fifty percent more efficient.
In this paper, we revisit the problem of the optimal summary statistics to compute the Stokes parameters from X-ray polarimetric observations. The paper is organized as follows. In Section 2 we start with the theoreticall
... (truncated, 31 KB total)00748aaee4f5b7c9 | Stable ID: Y2Q3ZmY2MT