Andi Peng announces the mission of Humans&, a company focused on building AI that centers human interaction, collaboration, and coordination. She emphasizes the need to move beyond traditional task-centric training approaches and scale toward more advanced capabilities.

The post highlights three key areas of innovation the team is pursuing: long-horizon memory, multi-agent systems, and user understanding. The tweet is truncated, suggesting additional details about their approach or vision were shared in the full thread.

This article by Erwin Lima explores a subtle but significant shift in how people use AI: the gradual, often unconscious transition from using it as a thinking partner (a "mirror") to treating it as a silent authority (an "oracle"). The author argues that the real risk of AI isn't misinformation or job displacement, but what he calls "authority drift" — the slow delegation of inner decision-making to AI systems. He illustrates this with two examples: an extreme case where someone stops consulting friends and mentors entirely, relying on AI for major life decisions, and a mundane case where AI progressively rewrites a person's professional identity until they can no longer distinguish their own voice from the optimized version.

The piece introduces a practical framework distinguishing Mirror Mode (using AI to surface patterns, contradictions, and blind spots while retaining decision authority) from Oracle Mode (effectively asking AI what to do, what to believe, or who to become). Lima emphasizes that the danger isn't in Oracle Mode existing, but in slipping into it unconsciously. The erosion manifests as less internal debate, faster but shallower decisions, and a gradual loss of authentic voice. His core message isn't anti-AI — it's a call for more conscious engagement, urging readers to use AI to sharpen their thinking while remembering that authorship over one's identity, narrative, and life direction must remain fundamentally human.

A research paper titled "Perfecting Human-AI Interaction at Clinical Scale" is shared by the Human-Computer Interaction Papers account on X. The paper focuses on turning production signals into safer, more human conversations, authored by Subhabrata Mukherjee, Markel Sanz Ausin, Kriti Aggarwal, Debajyoti Datta, Shanil Puri, and others. It spans multiple research areas including Human-Computer Interaction (cs.HC), Artificial Intelligence (cs.AI), Computational Linguistics (cs.CL), and Multiagent Systems (cs.MA).

The work appears to address the challenge of scaling human-AI conversational systems in clinical settings, leveraging real-world production data to improve both safety and naturalness of AI-driven interactions. This is a significant area of research as AI systems become more prevalent in healthcare and other high-stakes domains where the quality and reliability of human-AI dialogue is critical.

Lilian Weng shares a reflection on the intensive effort behind producing a training run log book, which resulted in 12 versions (plus many subversions) and 137 pages over the course of several months. She highlights the demanding yet enjoyable collaborative process involved.

Her key takeaway is that human-human collaboration turns out to be essential for improving human-AI collaboration — a notable insight from someone deeply embedded in AI research at OpenAI. The post underscores that behind advances in AI systems, significant teamwork and iterative human effort remain critical.

This literature review surveys the evolution of Human-AI interaction research, arguing that the field has been overly focused on Explainable AI (XAI) — making AI systems understandable to users — while neglecting more impactful forms of interaction that give users genuine agency. The authors systematically analyze studies across Human-Centered AI, Contestable AI, Interactive AI, and Interactive Machine Learning (IML), identifying a critical gap: most existing work treats humans as passive recipients of AI explanations rather than active participants in AI system design and evolution.

The review reveals several key findings. A wide range of studies fail to include end-users in co-designing AI systems, and recent research on active interaction tends to focus on low-risk domains like education and leisure while neglecting high-stakes areas such as healthcare and security. Most AI systems prioritize enhancing user experience for acceptance rather than enabling meaningful user control over AI mechanics. Only a small number of studies, primarily in the Interactive Machine Learning domain, allow users to actively modify AI internals — representing an underexplored opportunity for greater user empowerment.

The authors propose a framework centered on three dimensions — AI Users, AI Implementations, and AI Goals — and advocate for interactions that are explicit (obvious to the user), intentional (user-initiated), and informed (the user understands the system's functionality). They call for a shift from theoretical proposals to practical experimentation, offering design guidelines that promote participatory human-AI interaction and shared responsibility between users and AI systems. The work highlights that achieving truly interactive AI requires moving well beyond explainability toward systems where humans are co-creators, not just consumers.

This paper presents a large-scale dynamic experiment with over 800 participants across 40+ countries investigating how exposure to AI-generated ideas from ChatGPT affects human creativity, idea diversity, and the evolution of collective thought. Using a variant of the Alternate Uses Task (AUT), a standard divergent thinking measure, participants brainstormed creative uses for everyday objects after viewing example ideas that were either human-generated (from prior participants) or AI-generated. The researchers manipulated two key factors: AI exposure level (none, low, or high) and AI disclosure (whether AI-generated ideas were labeled as such). The experiment's dynamic design, inspired by the "multiple-worlds" paradigm, allowed ideas from earlier participants to feed forward as stimuli for later ones, simulating how AI content compounds through a cultural feedback loop.

The key finding is that high AI exposure increased collective idea diversity but did not improve individual creativity — in short, AI made ideas different, not better. High AI exposure also accelerated the rate of change in idea diversity over time, a compounding effect with potentially significant long-term implications. There were no main effects of disclosure (labeling ideas as AI-generated), though interesting moderating effects emerged: self-identified creative individuals were less influenced by AI disclosure, while participants appeared more likely to knowingly adopt AI ideas when the task was difficult. The study contributes a novel experimental framework for studying passive AI exposure — the increasingly common scenario where people encounter AI-generated content without actively soliciting it — and suggests that while AI integration may enrich the diversity of a collective idea pool, it does not necessarily elevate the quality of individual contributions.