iiinova — where your digital self keeps evolving

Dan Lametti

@daniellametti.com

1.8K followers 975 following 139 posts

Professor of Cognitive Psychology at Acadiau.ca. Psycholinguistics. Speech Processing. Human-AI interaction. Running. www.lamettilab.com Halifax, NS.

Posts Replies Media

Dan Lametti @daniellametti.com · 03/08/2026

New work led by @joannakuc.bsky.social: "Human or AI? An Interpretable Method for Auditing Free-Text Data in Online Psychological Research." The paper presents a method to identify AI-use in free text entry (e.g. journal entries, reflections) without using black box software osf.io/preprints/ps...

Figure 2. Linguistic markers distinguishing perceived human and AI authorship. Left panel: LASSO coefficients for word-based LIWC-22 features predicting perceived AI versus human authorship. Features associated with entries judged as human-written are represented by negative coefficients, while features associated with entries judged as AI-generated are represented by positive coefficients. Only features with non-zero coefficients after L1 regularisation are shown. Right panel: Wordclouds for the top three content LIWC categories per direction. The left column shows categories associated with entries judged as human-written: cognition, quantity, and leisure. The right column shows categories associated with entries judged as AI-generated: perception, memory, and auditory perception. Word size reflects frequency within each category.

022

Dan Lametti @daniellametti.com · 29/06/2026

New work led by @thelablab.bsky.social out in Neuroscience and Biobehavioral Reviews. "Inner Speech and the Neurobiology of Psychosis". Highlights below. www.sciencedirect.com/science/arti...

The image shows a bullet point list of highlights from the paper:
• First meta-analysis mapping inner-speech systems onto brain differences in psychosis
• Psychosis shows increased premotor activity, preferentially engaged by covert speech
• Psychosis shows reduced auditory and ACC grey matter, more engaged by overt speech
• Covert premotor and overt auditory regions form inversely coactive networks
• Results bridge predictive, salience, and self-monitoring accounts of hallucinations

1104

Dan Lametti @daniellametti.com · 06/06/2026

New preprint from my lab led by Acadia Master's student Lindsay Heyland. We provide evidence for a purely motor deficit in real-world L2 speech. "A Motor Basis for a Foreign Accent? Linguistic Context Constrains the Generalization of Second Language Speech Motor Plans". osf.io/preprints/ps...

Upper panel: Arrows reflect sensorimotor adaptation associated with bilingual sentence production—the average change in F1-F2 production in English (L1) and French (L2) for each participant in each phase of Experiment 1. The extent to which these changes opposed altered auditory feedback experienced in the adaptation phase is shown in the bars below. Lower panel: Bars reflect average direct adaptation, transfer of adaptation to part sentences and reversed part sentences, and after-effect magnitudes in L1 and L2.

Participants showed significantly more transfer of sensorimotor adaptation from sentences to part sentences in their first language (English) compared to their second (French). In English (L1), adapted motor plans transferred completely to part sentences; in French (L2), transfer was markedly weaker, despite the presence of robust L2 aftereffects when they returned to sentence production—the speaking context associated with adaptation. The results suggest that compared to their first language counterparts, second-language speech-motor representations are more closely tied to the linguistic context in which they are learned.

0102

Dan Lametti @daniellametti.com · 27/05/2026

New paper in the Journal of Psychopharmacology led by UCL PhD student @joannakuc.bsky.social and the @thelablab.bsky.social "Speech markers of psychological change following a psychedelic 5-MeO-DMT retreat". journals.sagepub.com/doi/10.1177/...

Following 5-MeO-DMT, speech markers revealed increased cognitive language, decreased social words, and altered voice quality

054

Dan Lametti @daniellametti.com · 20/08/2025

We added a preregistered study to our paper, Affiliation in Human-AI Interactions Based on Shared Psychological Traits. We created bespoke LLMs that exactly mirrored participants' personalities or the opposite. Participants greatly preferred chats with their mirror AI: osf.io/preprints/ps...

031

Dan Lametti @daniellametti.com · 23/04/2025

Exciting work from UCL PhD student Joanna Kuc and fellow @thelablab.bsky.social members: "Speech markers of psychedelic-induced psychological change". Preprint: www.biorxiv.org/content/10.1...

021

Dan Lametti @daniellametti.com · 06/03/2025

We have a new preprint on human-LLM social interaction: "Affiliation in human-AI interactions based on shared psychological traits". TLDR: People feel more similar to and understood by an LLM when it mimics aspects of their own psychology, such as anxiety and extroversion. osf.io/preprints/ps...

Title: "Affiliation in human-AI interactions based on shared psychological traits".

Abstract: People connect with others who share their likes, interests, and mental states, a phenomenon known as homophily. Here, we examine human-AI homophily by testing how individuals' psychological traits are related to their affiliation with large language models (LLMs) that mimic these traits. In Experiment 1, the LLM GPT-4 was prompted to produce language aligned with either an anxious or non-anxious state. Participants (n=100) engaged with both versions of the LLM and then completed a questionnaire to assess how relatable they found the AI. Participants with high anxiety felt more similar to and understood by the LLM instructed to mimic an anxious state, while participants with lower anxiety reported a perceived difference between themselves and the LLM. In addition, the sentiment of the messages exchanged with each LLM persona mirrored the sentiment of the messages the LLM produced. This apparent display of empathy toward the AI was predicted by participants’ own anxiety in the case of messages with a negative sentiment. In Experiment 2, participants (n=100) engaged with two LLMs prompted to align with either an extroverted or introverted personality. Extroverted participants felt similar to the AI when it mimicked extroversion. Taken together, the findings support the idea that humans affiliate more with LLMs when they are instructed to mirror aspects of their psychology.

021

Dan Lametti @daniellametti.com · 20/02/2025

In the Corpus Christi College Oxford alumni magazine, there's a nice profile of our work with @robinmurphy.bsky.social's research group and OneReach.ai on human-AI social interaction (paper imminent!). I spent the fall at Corpus—lovely place! www.ccc.ox.ac.uk/sites/defaul...

020

Dan Lametti @daniellametti.com · 10/12/2024

Every now and then IG ads serve up a useful purchase. Archie loves his cat hammock.

240

Dan Lametti @daniellametti.com · 23/11/2024

My favourite Oxford college garden, because it's beautiful and they actually encourage you to walk on the grass.

The garden of Corpus Christi College Oxford on a sunny day in November.

1100

Dan Lametti @daniellametti.com · 15/09/2024

GPT o1 "thinking" a bit too hard about this one ...

000

Dan Lametti @daniellametti.com · 31/08/2024

Happening Tuesday. AI: How it works and how to use it responsibly in university. Open to all Acadia students and the world via the livestream.

011

Dan Lametti @daniellametti.com · 18/01/2024

TT position in Developmental Psychology (broadly defined) at Acadia. Come join us! tiny.cc/js97wz Good salary, benefits (e.g., defined benefit pension), and working conditions thanks to our incredible union. Great, supportive department!

0139