Article
CommunityJev Benchmarks & Evaluations
Nudging Decision Models: Do Digital Nudges Steer Fast, Generation-Free AI Agents?
arXiv study adapting an online shopping experiment to three decision models, including Jev, to test whether digital nudges steer non-generative System One models. Reports 600 runs per model and compares default and social-influence choices against a control.
- Format
- Article
- Reading time
- 2 min
- Published by
- Community
- Added
- Nov 2026
Highlights
- Adapts an online shopping experiment previously run with humans and LLM agents to three decision models.
- Tests Jev, Decision-1, and d1 across 600 runs per model.
- Models chose a preselected default in 97.5-99.9% of choices.
- Social influence information shifted choices to 75-96%, versus 47-49% in the control.
- All three responded more strongly to the automatic nudge, opposite to reasoning-capable LLM agents.
Questions about this entry
Short answers drawn from the linked project or page.
It asks whether non-generative decision models can be steered by digital nudges online, and whether automatic nudges affect them more than reflective ones, as dual process theory predicts.
Three decision models: Jev, Decision-1, and d1. The experiment was adapted from an online shopping study previously run with humans and LLM agents.
600 runs per model were collected, giving the study its comparison base for default choices and social influence effects.
Across the 600 runs per model, the models chose a preselected default in 97.5-99.9% of choices.
A product with social influence information was chosen in 75-96% of choices, compared with 47-49% in the control condition.
All three models responded more strongly to the automatic nudge, the opposite of what has been reported for reasoning-capable LLM agents.
Decision models are a class that chooses among options without generating text; developers call them System One models, referencing dual process theory, and describe them as faster and less costly than LLM agents.
More Jev resources like this
arXiv study that evaluates ride-hailing rebalancing policies under the EU Platform Work Directive's disclosure requirement, using self-interested LLM driver agents built on Claude Haiku 5.5 and Jev in a three-zone model and in Manhattan.
Jev Benchmarks & EvaluationsArticle#arxiv#research#agents
Community
arXiv paper introducing S²-MAS, a society-scale multi-agent simulation framework built on 500 million synthetic household personas from survey data. Its Genese engine separates decisions from state updates, and sharp-421m-mlx is benchmarked against Jev 1.13 on financial-choice tasks.
Jev Benchmarks & EvaluationsArticle#arxiv#evaluation#benchmark
Community
A benchmark evaluating Theory of Mind selection in typed decision models against generative baselines: 11,117 English questions from nine ToM benchmarks across six capability groups. TypeSafe Jev reaches 73.2% source-balanced accuracy and ranks first or second on every capability.
Jev Benchmarks & EvaluationsArticle#community#research#arxiv
Community
HKUST-KnowComp/Jev-ToMPython
HKUST-KnowComp/Jev-ToMPythonAn empirical comparison of Jev as a typed judge for retrospective agent-trace security against four generative judges on 5,219 trajectories. Jev averages positive-class F1 77.8 versus 74.1 for GLM-5.2, leading on two of four benchmark collections.
Jev Benchmarks & EvaluationsArticle#community#research#arxiv
Community
A paired, self-audited evaluation of an open-weight (Laya) and a hosted (Jev) System-1 decision model on 11 agent decision points: 7,283 base cases plus 6,640 robustness variants. Jev is significantly more accurate on 9 of 11 decision points, by 10.8 to 46.0 points.
Jev Benchmarks & EvaluationsArticle#community#research#arxiv
Community
David-DL-Space/sys1-evalJavaScript
David-DL-Space/sys1-evalJavaScriptAn empirical study of whether Jev's decision-level savings transfer to task-level gains, across five agents in ALFWorld and AndroidWorld. Across 4,500 task executions, decision cost and latency fall in all five agents, but success declines in two mobile agents and per-outcome cost rises in one.
Jev Benchmarks & EvaluationsArticle#community#research#arxiv
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.