AI

Meet your research partner for AI’s hardest questions at NeurIPS

Data on what people think, do and decide, from verified participants who are right for the question. Ready for your next release review, paper or policy decision.

Trusted by researchers at the frontier

Prolific at NeurIPS 2026

Make room for a better conversation.

Bring the study you're planning.
20 minutes with our team on rater selection, repeat panels or your human feedback workflow. We'll bring study data matched to your problem.
Book 20 minutes in Sydney
Watch an agent launch a study.
At booth 118, an agent sets prescreeners, picks a returning panel and launches a preference evaluation through our MCP server, then the quality signals come back live.
Find us at booth #118
Rooftop Welcome Party
​Join fellow AI researchers and technical leaders for an evening of food, drinks, and skyline views at an exclusive rooftop party in Sydney.
Request a place

Our research at NeurIPS

ACCEPTED TO NEURIPS 2026

MetaLoop: Benchmarking the Full Metacognitive Loop in LLMs

Thursday, Dec 10 at 17:00 in Hall 1-4 (Sydney Poster Session 6)

A seven-task benchmark that tests how models monitor uncertainty, act on it and learn from feedback. The research compares 12 frontier models with 460 human participants—and finds that accuracy does not predict metacognitive ability.

Nora Petrova, John Burden and Jerome Wynne

Learn more
RESEARCH COLLABORATION

Theory of Mind and Persuasion Beyond Conversation

Tuesday, Dec 8 at 10:00 in Hall 1-4 (Sydney Poster Session 1)

Can a model shape another agent’s beliefs through planning and action? This work tests social reasoning beyond dialogue, with implications for how we evaluate autonomous agents.

Led by Ben Slater, with Matteo G. Mecattaf, Lucy G. Cheke, John Burden and Winnie Street.

Learn more
BiAlign workshop - NeurIPS 2026

NeurIPS 2026 Workshop: Dynamic Alignment in Human–AI Coupled Systems

Friday, Dec 11 at 10:55 in Cockle Bay Room 1

Our team has an accepted paper at the BiAlign workshop. Come along to explore the research, bring your questions and join the discussion in Sydney.

More from the NeurIPS community.

Accepted at NeurIPS 2026, these papers cite Prolific and were authored by independent research teams outside Prolific.

FRAUDBENCH: A Legal Evaluation of AI Deception
The evaluation contains a novel dataset of 159 scenarios based on real-world U.S. court cases and validated by legal experts.
Read the paper
An Enigma of Artificial Reason
Can a model solve a problem but fail to spot flawed reasoning? This study separates getting the answer right from judging the steps behind it, exposing an answer-confirmation bias in reasoning models.

Authors: Mingzhong Sun, Teresa Yeo, Armando Solar-Lezama & Tan Zhi-Xuan.

Institutions: National University of Singapore (NUS), Massachusetts Institute of Technology (MIT), Singapore-MIT Alliance for Research and Technology (SMART), A*STAR.
Read the paper
Political Neutrality as Balanced Approval: A Large-Scale Human Evaluation of AI Responses
What would politically neutral AI look like to people who disagree? The PARETO study proposes balancing approval across opposing viewpoints and tests that definition across 20 contested US issues.

Authors: Jonathan Stray, David Zhai Yang, Steven Luo, Miu Nicole Takagi & Serina Chang.

Institutions: University of California at Berkeley, Amazon Japan.
Read the paper
When Empathy Misses the Goal: A Benchmark for Goal Displacement in LLM Advice
Examines whether reassuring, empathetic advice actually advances the user’s goal. Evaluation insight: measure goal fidelity separately from warmth or perceived helpfulness. Particularly relevant to assistants and domain-specific AI.

Authors: Dean Ariel, Guy Laban

Institutions: Ben-Gurion University, Tel Aviv University and Clalit Health Services
Read the paper
REALITYTEST: How People Probe AI Identity and Whether Models Disclose It
Tests whether models disclose their AI identity when questioned in ways real people actually use. Evaluation insight: realistic human queries can expose behaviours that narrow or synthetic tests miss. Strong relevance to frontier AI, trust and policy.

Authors: Anna Gausen, Sarenne Wallbridge, Bessie O'Dell, Christopher Summerfield, Hannah Rose Kirk

Institutions: UK AI Security Institute
Read the paper

Working through a complex research question?

What are you researching next?

Tell us what you're evaluating. We'll come back with a proposal for how we'd design and deliver it.