conspiracy theory-type rabbit-holes, where a lot of the online sources are fringe and hotly debated, it's easier to believe what All-Powerful-And-Knowing Chatbot says. Safeguards related to AI literacy could help.

I also really dislike his stigmatizing use of the word "crazy"

Agreed, it's odd he implies that psychosis is either genetic or environmental. Most of this boils down to a diathesis-stress model, with AI-provided info as the stressor and, as @dingdingpeng.the100.ci mentions, dynamics. Big issue with AI is the apparent trustworthiness of the source. Cf...

The causality facade isn't limited to AI/ML, of course. But the more complex and black-boxy the model, the easier that facade is to ignore/downplay.

difficult to validate and built on untestable assumptions about the causal structure, confounds etc. Adding more covariates doesn't necessarily help. See "The Good, the Bad, and the Ugly" section on CausalML here arxiv.org/pdf/2206.15475

Big thanks to @margotvdweijer.bsky.social for spearheading this effort! And to our coauthors @appelbap.bsky.social, Emily Bassett, and @lucasjmatthews.bsky.social

S/O to my advisors, @jorsmo.bsky.social and Ben Neale; our wonderful collaborators in Finland and @fimm-uh.bsky.social, especially Aarno Palotie and Olli Pietilänen; and the Stanley Center for Psychiatric Research @broadinstitute.org

Global functioning is used widely to inform treatment and prognosis. In our sample, GF scores were associated with important severity- and course-related variables including hospitalization burden and cognitive performance.

We addressed those gaps using the SUPER-Finland study, which includes linked register data for every single psychiatric hospitalization in Finland since 1969. Global functioning has been assessed routinely at inpatient admission and discharge nationwide since 1994. A peek at the raw GF scores:

Some background: recently, interest has increased in using PGS to predict not only lifetime SZ risk, but also clinically relevant disease outcomes. Studies have been limited by small Ns and superficial phenotypes. Here's a nice review of that literature: www.tandfonline.com/doi/full/10....

These results suggest that integrating genomics with clinically relevant, longitudinal phenotypes may help parse heterogeneity in schizophrenia severity, prognosis, and course.

Caveat: as is typical in Psych Gen, effect sizes were small and PGS are not sufficiently predictive for clinical use.

Leveraging ~120k GF measures from a complete hospitalization register, we found that higher SZ PGS predicted worse GF at admission & discharge, and less functional improvement during hospitalization. Higher EA PGS predicted better discharge GF and greater improvement, but worse admit GF

Chapter 7 of Groth-Marnat, Gary (2016). Handbook of Psychological Assessment, Sixth Edition: Wiley.

Geared towards clinicians but provides a thorough history of MMPI development, updates, and psychometrics. PDF is available online!