TBNAI
The psychophysics of affect in large language models
Publication: cs.AI updates on arXiv.org
Author: Joseph Marks (NatWest Group, London, UK), Marius C. Vollberg (University of Zurich, Zurich, Switzerland, Swiss Center for Affective Sciences, Geneva, Switzerland)
Source published: 02/11/2026, 05:00:00 UTC
Briefing created: 02/11/2026, 05:24:18 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 02/11/2026, 05:03:25 UTC.
Briefings
A 2026 study by researchers testing Large Language Models (LLMs) on gambling and stock-trading outcomes found that LLMs exhibit silicon gain amplification, rating gains more intensely than losses. When outcomes are compared together, these ratings linearize, and in actual choice tasks, the models' own ratings do not better predict their decisions than standard monetary utility.
Key points
- LLMs rated gains more intensely than equivalent losses, an inversion of canonical loss aversion termed silicon gain amplification.
- In choice tasks, LLMs' own ratings failed to explain their choices better than monetary utility did.
Context
This research investigates whether agentic Large Language Models incorporate anticipated feelings into decision workflows, revealing malleable mappings from outcomes to affect.
Limitations
The study is a preprint (arXiv:2610.40985v1) with limited coverage of the full research scope. Results are observational and do not include causal claims or forecasts about future model behavior.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 02/11/2026, 05:24:18 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeiesgsoz2s37dmr7i7cjv3ckgfb2bjrc4jof4ab5iffb4jvvhsq6pa