Search

[2610.24520] HABIT-Bench: Evaluating Habit Induction from Longitudinal Weak Evidence in Agent Memory

HABIT-Bench assesses how well agent memory systems infer and apply user habits from weak, longitudinal evidence. Using synthetic histories across multiple domains and four-choice tasks, it finds a persistent gap between access to past interactions and accurate habit inference, with limited gains even when evidence is annotated.

Beyond Qualifications: Understanding People As AI Changes Work

The document argues that behavioral intelligence — how people approach tasks, communicate, and adapt — is crucial alongside qualifications for hiring, development, and leadership as roles evolve, especially in the age of AI-driven change.

[2611.03980] Measuring Model-Specific Alignment Drift in LLM-Based Opinion Dynamics

The study tests whether LLM agents’ opinions in repeated discussions drift from their initial stances due to interactions, using a fixed psychometric measure. It finds model-specific alignment drift, with some models preserving clusters under certain attentional setup, and highlights the importance of LLM choice as a validity factor in social simulations.

Martin Cordsmeier on X: "AI can write code, pass the bar exam and sound human. It still can’t tell when you’re unsure, holding back or about to walk away. Today we’re introducing Thorsten-4, a Large Psychology Model that predicts what people will do next." / X

Thorsten-4 analyzes communication to predict 50 behavioral and psychological traits, including motivations, decisions, and reactions under pressure. The text reports 90% predictive accuracy, checked against established assessments and real behavior.

The NeuroSoft | EEG & fNIRS Data Analysis

The article argues that AI profiles should reflect people’s capacity to change, rather than allowing past behavior to shape their future choices. It connects this concern to neuroplasticity and a proposed, untested hypothesis about the brain’s capacity for later remodeling.

SUS-Bench: the #1 AI EQ leaderboard

Research on social deduction games has relied on simulations, small studies, and recorded games, which may not reflect varied, natural human behavior. The text calls for real-time human-AI play at scale, with people playing for fun.

Artificial intelligence and biomarker revolution in psychiatry - Health News - Başka Gazete

Translated title. Original: Psikiyatride yapay zeka ve biyobelirteç devrimi - Sağlık Haberleri - Başka Gazete

The article reports a shift toward precision psychiatry using AI, digital phenotyping, and biomarkers to tailor treatments, reduce trial-and-error in care, and incorporate environmental factors alongside biology. It discusses digital data from smartphones and wearables, biomarker-driven predictions, and evolving DSM considerations to improve outcomes in mental health care.

Future Anxiety ⑤: The "improvement perspective" I felt was more important than the amount saved | Tanuki-san / AI Work Techniques Self-Improvement Financial Literacy Studying Money

Translated title. Original: 将来への不安⑤： 貯金額より重要だと感じた「改善の視点」｜たぬきさん/ AI 仕事術 自己啓発 金融リテラシー お金の勉強

The author argues that money anxiety is less about the absolute amount saved and more about having an explainable, improvable view of finances. By treating money management like a problem-solving process—identifying causes, breaking issues into concrete steps, and continuously adjusting plans—the author reduces anxiety and views finances as part of overall life planning.

AI-manipulated images can create false memories in people - Altered photos and videos double errors in memory tests - scinexx.de

Translated title. Original: KI-manipulierte Bilder können bei Menschen falsche Erinnerungen erzeugen - Veränderte Fotos und Videos verdoppeln Fehler im Gedächtnistest - scinexx.de

A study shows that AI-altered photos and videos can distort people's memories, with manipulated images causing more false recollections than unchanged images in a controlled MIT experiment involving 200 adults who viewed 24 photos and were split into four groups.

Can Artificial Intelligence Detect Lies in 10 Questions? - Onedio

Translated title. Original: 10 Soruda Yapay Zeka Yalanı Ayırt Edebilir mi? - Onedio

AI can flag patterns in voice, facial expressions, speech, and behavior that may be associated with lying, but these signs can have other causes. The article says AI can estimate the likelihood of deception, not prove it, so its findings should not replace human judgment.