INSIGHTS / RESEARCH NOTES
We document research on AI, data, behavioral science, and organizations, including methods worth testing in practice. Each note shares not only conclusions, but also hypotheses, observations, and limitations.
69 RESEARCH NOTES
2 / 6 PAGE
Based on meta-analysis corrections, re-analysis of publication bias, and large-scale administrative nudge trials, we shift the decision framework to prioritize baseline rates, absolute differences, heterogeneity, and costs in advance.
Read article
We distinguish between load reduction via external memory and the reduction of internal encoding and retrieval, measuring both the performance when AI is usable and the independent execution when it is not.
Read article
From experiments showing that evaluations change when a work is labeled as AI-generated, we separate blind content evaluation from eligibility reviews of provenance, rights, and production process.
Read article
We design procedures that do not equate fairness with merely displaying explanations, but instead enable individuals to understand, present their circumstances, correct data, and proceed to human review and relief.
Read article
Evidence that ranking methods yield inconsistent attitudinal effects despite altering exposure and usage necessitates the creation of a ranked evaluation separating information diversity, reliability, time utilization, and well-being.
Read article
Distinguish between connection immediately after conversation, loneliness after repeated use, comparison with humans, and relationship replacement; do not extrapolate short-term effects to long-term well-being.
Read article
Based on experiments showing that avoidance after observing algorithmic errors and acceptance with minor correction rights vary, we design operations that are neither mandatory use nor laissez-faire.
Read article
Do not treat names, voices, appearances, and emotional expressions as uniform trust-enhancing measures; design them based on context, user emotions, understanding of capabilities, expectation violations, and handover to humans.
Read article
We mitigate the risk of malicious documents entering the top search results and guiding answers through write history, index monitoring, contradiction detection, post-generation verification, and isolation with re-indexing.
Read article
Do not trust multi-LLM discussions based on vote counts; instead, determine the added value by assessing the diversity of initial answers, conformity, problem drift, retention of correct answers, and the difference from simple voting.
Read article
We do not judge success solely by the target correct answer for knowledge editing; instead, we measure paraphrase, locality, general capabilities, and forgetting after continuous editing to establish a boundary for switching to RAG or retraining.
Read article
Do not treat the completion of a deletion API as proof of forgetting; instead, comprehensively verify target memory, retention utility, leakage, and continuous deletion by using a retrained model as the baseline.
Read article
