There is a strange convergence happening in the psychology literature right now. In one paper, a mathematician-philosopher argues that the glowing vectors inside a language model are not, and have never been, a window into awareness — they are merely coordinates in a fixed basis, no more conscious than a spreadsheet cell. In another, a team in Santiago de Chile measures how many hours a six-year-old sleeps and how many caffeinated sodas she drinks before bed. In a third, a Brazilian outpatient clinic reports that an antidepressant, repurposed, may finally ease the bone-deep exhaustion that millions of Long COVID patients still carry. And in a fourth, a dozen neuropsychologists in North America sit down to write the rules for how the next generation of their field will be trained.
Taken together, these five papers — published between October 3 and 10, 2026 — sketch a field in the middle of a quiet but consequential reorientation. Psychology is no longer just studying the mind in isolation. It is defining what the mind is against (AI systems that mimic it), what the mind does to the body (sleep, fatigue, the long tail of viral infection), and how the mind's study is instituted (training, guidelines, the architecture of a profession). The common thread is boundary-work: drawing lines between what we know, what we can measure, and what we are tempted to claim.
The Vector That Does Not Think
The most philosophically charged paper of the week is Syed Muntasir Mamun's Contextual Vectors Are Not Awareness [5]. The argument is deceptively simple once you see it. Contemporary language models assign a vector to each token, and that vector shifts depending on the surrounding context. A growing literature — Mamun calls it the "consciousness spectrum" literature — reads the geometry of these vectors as evidence of something like situational awareness, even a dim form of experience. Mamun's intervention is to apply a criterion from tensor mathematics: a multidimensional array only becomes the component representation of a geometric object when a transformation law is stated and an invariant is named. Without those, it is just a table of numbers in a basis that was fixed by training.
He then separates three things that the consciousness literature tends to collapse into one. First, the representation map: a function from contexts to a fibre, which is genuine mathematics of the vector. Second, a self-locating competence — the ability to reason out of context, measured behaviorally by tasks like those in the Situational Awareness Dataset. This is a family of tasks, not a property of the vector. Third, an identity claim, the kind associated with integrated information theory, on which experience is a cause-effect structure. Mamun's point is that no embedding in high-dimensional space entails the third. The vector is the first. The tasks are the second. The metaphysics is the third. Conflating them, he argues, is a category error dressed up as a discovery.
The constructive part of the paper reframes context as a base, the contextual vector as a coordinate representation of a section, and situational content as an invariant functional of that section — something stable under the redescriptions the theory itself permits. He notes that current residual streams in transformer architectures fail this test: they are written in a basis fixed by training, and a change of basis is not part of the model's operations. Multi-scale context, from the token to the deployment regime, is where the real difficulty lies, and where "fractal and fractional warnings apply." A single vector in a fixed dimension collapses scale. As Mamun puts it, the isolation "decides only which mathematical object a claim has in fact produced" — it does not, and cannot, settle the hard problem of consciousness.
Mapping the Human-AI Boundary
If Mamun's paper is the philosophical scalpel, Andreas Ehstand's NEOMANITAI framework [1] is the cartographer's grid. The paper develops a faceted research framework for describing human-AI interaction across digital, embodied, and collective settings. It is explicitly a conceptual working paper — no new participant data, no empirical validation — and Ehstand is careful to say its usefulness "remains to be tested against simpler descriptions and existing approaches." What it offers is an inspectable connection between programme navigation, episode descriptions, and claim-level provenance. Eight ordering conventions, seven descriptive facets, a small typed relation model, three constructed civilian cases, and four comparative research questions make up the skeleton.
The framework builds on a targeted literature review spanning distributed cognition, socio-technical systems, hybrid intelligence, collective intelligence, human-robot interaction, and taxonomy development. Its ambition is not to explain how humans and AI interact, but to give the field a shared vocabulary for describing those interactions at different levels of granularity — from a single episode of a person using a chatbot to a collective deployment regime in which multiple agents and humans coordinate. The constructed cases are explanatory, not empirical. The paper is, in a sense, the infrastructure that would allow a field like the one Mamun is critiquing to state its claims with the precision the mathematics demands.
There is a productive tension between these two papers. Mamun says: stop reading consciousness into vectors. Ehstand says: let's build a framework so precise that we can say exactly what a human-AI interaction is, at which level, under which conditions. One is a warning against overclaim; the other is a tool for underclaiming less. Together, they suggest that the psychology of human-AI interaction is entering a phase where the field must define its objects of study before it can study them.
The Body Keeps the Score: Sleep, Caffeine, and the Chilean Schoolyard
Step away from the vectors and the tensor algebra, and the psychology of the body looks very different. Durán-Agüero, Cediel Giraldo, and Brignardello Guerra surveyed 805 school-age children (6 to 10 years old) across six neighborhoods in Santiago, Chile, asking parents about sleep duration, physical activity, and food intake, while measuring the children's anthropometrics [2]. The results are stark. 52.6% of the children were obese. Only 46.4% slept the recommended ten or more hours. Normal-weight children slept significantly more than obese peers — 9.8 ± 0.9 hours versus 9.6 ± 0.9 — and sleep duration during the week was inversely associated with obesity (OR: 3.5, 95% CI 1.3–9.2).
But the finding that may matter most for the field's understanding of developmental psychophysiology is the caffeine data. At night, 52.2% of children consumed caffeinated soft drinks, 32.6% drank coffee and/or tea, and 21.2% consumed both. This is not a marginal habit; it is a near-universal pattern in a sample where more than half the children are already obese. The paper does not isolate caffeine as the causal driver of shorter sleep or higher weight — the design is cross-sectional and correlational — but it places the body, its chemistry, and the social environment of the schoolyard in a single frame. In a field that has spent decades parsing cognition, attention, and executive function, this is a reminder that the developmental substrate is, in many cases, a six-year-old drinking a caffeinated soda before bed.
When the Mind Will Not Rest: Fluvoxamine and the Long Tail of Fatigue
If the Chilean study captures the body in its everyday context, the Brazilian trial by Reis and colleagues captures it in its most intractable form: the fatigue that lingers 90 days to 24 months after SARS-CoV-2 infection [4]. The randomized, placebo-controlled, adaptive trial enrolled 399 adults across outpatient sites in Brazil and tested two interventions — fluvoxamine and metformin — against a matching placebo, each for 60 days with follow-up to day 90. The primary outcome was change in the Fatigue Severity Scale (FSS) score.
The headline result: fluvoxamine showed a mean difference of −0.43 at day 60 and −0.58 at day 90 versus placebo, with a posterior probability of superiority of 99% at both time points. It also improved quality-of-life scores with a high posterior probability of superiority. The metformin arm, by contrast, was inconclusive — a dose change recommended by the data safety monitoring committee due to adverse effects led to early termination of that group in September 2024. Adverse event rates were lower for fluvoxamine (24.0%) than for either metformin dose (42.5–45.1%) or placebo (35.5%), and Grade 3 or higher adverse events were rare (≤2.7%), with no treatment-related deaths.
Several caveats are important and the authors state them plainly. The benefit of fluvoxamine beyond 90 days is unknown. The trial did not address other long COVID symptoms. And the metformin result cannot be read as a null finding — it is a result of a broken experiment, not a failed drug. What the trial does establish, with the statistical confidence of a Bayesian adaptive design, is that a serotonergic agent can move the needle on the single most common and debilitating symptom in a condition that has, until now, had almost no targeted pharmacotherapy. For the psychology of chronic fatigue — a field that has long struggled to distinguish central from peripheral mechanisms, motivational from physiological causes — this is a data point that forces a reckoning with the body's pharmacology as a legitimate and necessary part of the therapeutic picture.
Building the Profession: The NAPSN Guidelines and the Architecture of Training
The fifth paper of the week is the most institutional, and perhaps the most quietly consequential. Soble, Towns, Silva, Newman, González, Abrams-Silva, and colleagues document the development of the North American Association of Practicum Sites in Neuropsychology (NAPSN) organizational training guidelines [3]. The NAPSN was established in 2023 to support external practicum programs across urban, suburban, and rural settings in the United States and Canada. A subcommittee of six practicing clinical neuropsychologists — spanning adult, pediatric, and lifespan populations; academic medical centers, the VA, hospital-based, and private practice settings; and clinical, forensic, and research orientations — developed four core content areas through consensus: supervised clinical practice, didactics and educational programming, professionalization, and research.
The guidelines are explicitly not prescriptive rules. They are designed to function as supportive scaffolding, allowing maximum flexibility to account for the range of existing practicum settings. This is a deliberate choice in a field where a rural VA hospital's practicum looks nothing like a private forensic neuropsychology practice in a major city, and where the old model of a single, uniform curriculum has long been inadequate. The process — multiple internal revision rounds, presentation to the NAPSN Board of Directors and Elected Officers, further revision — mirrors the kind of iterative, consensus-driven knowledge production that the field values, even when the output is a set of guidelines rather than a hypothesis test.
What makes this paper matter beyond its immediate audience is that it is an act of field-formation. It codifies, in a public document, what a neuropsychology practicum is supposed to be, across a continent, in a profession that has historically relied on apprenticeship and local norms. In a week when another paper is arguing about whether a vector in a language model constitutes awareness, the NAPSN guidelines are a reminder that the science of the mind is also a human institution, with training pipelines, supervisory relationships, and the unglamorous but essential work of making sure a fourth-year doctoral student in a small-town clinic gets the same quality of education as one in a university lab.
The Bigger Picture: A Field Defining Its Edges
What do these five papers have to do with each other? On the surface, almost nothing. A tensor-theoretic argument, a Chilean school survey, a Brazilian pharmacology trial, a training guidelines document, and a taxonomy for human-AI interaction do not form a single narrative. But they share a deeper impulse: the need to define the object before you study it. Mamun defines what a vector is and is not. Ehstand defines the facets along which a human-AI interaction can be described. Durán-Agüero and colleagues define the measurable variables of a child's sleep and nutrition in a specific urban context. Reis and colleagues define the primary outcome (FSS change) and the boundaries of a trial that was, in part, broken by its own safety monitoring. Soble and colleagues define what a practicum in neuropsychology should contain, and what it should not be forced to contain.
The through-line is epistemic humility paired with methodological rigor. Each paper, in its own register, resists the temptation to overclaim. The vector is not a mind. The cross-sectional survey does not prove that caffeine causes obesity. The fluvoxamine result does not extend beyond 90 days or to other long COVID symptoms. The training guidelines are not a law. The NEOMANITAI framework has not yet been tested against simpler descriptions.
And yet, in that restraint, there is a kind of confidence. The field is large enough, now, to hold these conversations simultaneously — the philosophical, the clinical, the developmental, the institutional. Psychology in October 2026 is not one story. It is a constellation of stories, each one drawing a slightly different line around what the mind is, what the body does, and what it means to train someone to study either. The vectors are not aware. The children are not sleeping enough. The fatigue is real, and for the first time, a pill might help. And somewhere in a suburban VA hospital, a practicum student is learning to read a neuropsychological battery under the supervision of a clinician who followed guidelines that did not exist three years ago.
That is the field. Messy, multi-scaled, and slowly learning to say precisely what it means when it says the word mind.
References
- Andreas Ehstand (2026). NEOMANITAI: A Faceted Research Framework for Human-AI Interaction. Zenodo (CERN European Organization for Nuclear Research).
- Samuel Durán‐Agüero, Gustavo Andres Cediel Giraldo, Jerusa Brignardello Guerra (2026). Relationship between nutritional status and sleep duration in Chilean school-age children. Archivos Latinoamericanos de Nutrición.
- Jason R. Soble, Stephanie J. Towns, Marc A. Silva et al. (2026). Facilitating best practices in practicum training in clinical neuropsychology: introducing the North American Association of practicum sites in neuropsychology (NAPSN) organizational training guidelines. Journal of Clinical and Experimental Neuropsychology.
- Gilmar Reis, Eduardo Augusto dos Santos Moreira Silva, Daniela Carla Medeiros Silva et al. (2026). Effect of Fluvoxamine and Metformin for Fatigue in Patients With Long COVID. Annals of Internal Medicine.
- Syed Muntasir Mamun (2026). Contextual Vectors Are Not Awareness: Reinterpreting Representation Geometry after the Tensor Test. Zenodo (CERN European Organization for Nuclear Research).