Nick Byrd, Ph.D.
LightNews LightNews
banner
byrdnick.com
Nick Byrd, Ph.D.
@byrdnick.com
3.3K followers 530 following 14K posts
I study how to improve decisions and well-being at @GeisingerCollege.bsky.social.

🎓 gScholar: shorturl.at/uBDPW

▶️ youtube.com/@ByrdNick

👨‍💻 psychologytoday.com/us/blog/upon-reflection

📓 byrdnick.com/blog

🎙️ byrdnick.com/pod
Posts Replies Media Videos
Pinned
byrdnick.com
Nick Byrd, Ph.D. @byrdnick.com · Oct 9
The #AI companies are onto me!

#OpenAI recently automated the identification of tasks worthy of "slow" #reasoning.

In "Strategic Reflectivism...", I showed why that's a key to #intelligence (in humans as well).

The #preprint (accepted in #LNCS) is now available as an audiopaper (a.k.a. #podcast)👇
Upon Reflection, Ep. 16: Strategic Reflectivism | Nick Byrd, Ph.D.
In late 2025, artificial intelligence companies like OpenAI popularized the idea of automating the process of selecting which model is best for a task. This allowed users to simply send their promp…
byrdnick.com
byrdnick.com
😳 Cohen's d of 1232?

Ooooh. The decimal point is missing. An f of 0.616 can transforms to 1.232 (www.psychometrica.de/effect_size....). 😌

Source: doi.org/10.1007/s111...
"Although only two or three words per text were changed (in texts with an average length of 112 words), these changes had a large effect (f = 0.616), which, converted into Cohen’s d, corresponds to d = 1232." Converting an f of 0.616 to a d results in 1.232.
January 15, 2026 at 3:52 PM
byrdnick.com
The more I learn about the status quo of #education assessment in #higherEd, the more I find opportunities to improve.

This paper reports similar opportunities in #MedEd: jamanetwork.com/jour...

How can a #university or #medSchool better embrace #learningScience system-wide?
“learning health systems [embed] research… in clinical operations to generate knowledge alongside care. We propose developing analogous learning medical schools, where systematic experimentation is a core component of curricular development.” “…learning medical schools could regularly exploit natural experiments in medical education. Schools often assign students to different clinical training sites or campuses, clerkship sequences, and preclerkship learning groups. These assignments typically are, or could be, random. Do students with earlier primary care rotations more frequently pursue primary care? Do students placed at higher-volume clinical sites achieve higher board scores or residency competency ratings? Leveraging inherent randomization… could answer such questions with minimal added cost or infrastructure. “More ambitiously, institutions could commit to systematic curricular experimentation…. randomize students to different experiences …and evaluate outcomes.”
January 15, 2026 at 12:23 PM
byrdnick.com
"it appears that Edgeworth was approximately 134 years ahead of this time"

He proposed the p < .005 criterion back in 1885, before the 2018 paper from Benjamin et al.

Via E.J. Wagenmakers on #Bayesian Spectacles blog.

Thanks to @jaspstats.bsky.social on #Mastodon!
Redefine Statistical Significance Part XXI: Edgeworth Proposed the .005 Criterion Back in 1885
The statistical significance test was not invented by Ronald Fisher. The key idea was already laid out by Francis Ysidro Edgeworth (1845-1926), whose 1885 article “Methods of statistics&#8221…
www.bayesianspectacles.org
January 14, 2026 at 2:10 PM
byrdnick.com
Do #AI "reasoning" models think reflectively?

Prompting #LLMs to reflective on and make a plan to correct their faulty intuitions on reflection tests resulted in "chance" improvement, lacking "principled application of reason".

"Illusions of reflection": doi.org/10.48550/arX...
Agent protocol (single LLM session) 1. Baseline answer. Model receives the task prompt and outputs four candidate items. 2. Self-reflection. For each failed item (detected automatically; rubric below) the same model is asked to explain the failure and produce a short corrective advice block (keywords, explanation, step-by-step instructions). 3. Re-answer. This advice block/reflection text is prepended to the original task prompt reminding of the constraints and the model attempts the task again. PassRate initial = Correct initial / Total initial PassRate reflection = (Correct initial + Correct reflection) / Total initial "reflection refers to its subsequent *reflection-based* re-attempts." "Reflection repeats the original failure above chance. Across reflection attempts, the repeat-category rate was 85.36% (567 attempts) when pooling across tasks and models. Relative to a within-cell category-mix benchmark, repetition was higher: observed 85.36% vs. benchmark 74.69% (95% permutation interval 72.13–77.25), an excess of +10.68 pp; permutation p = 0.0001 (Methods/Supplementary). Strategy terms were not reliably different in the pooled model (all p ≥ .30)." "Contrary to expectation, LRMs marketed for extended reasoning showed no reflection advantage over non-reasoning peers after multiplicity control." "Even when scores improve, the gains are not consistent with principled diagnosis and systematic correction of specific errors. Instead, it appears to be a chance event; resulting from having another attempt and occasionally producing valid items among continued failures. This is not indicative of a mechanism that identifies faults and prevents reoccurrence through principled application of reason."
January 14, 2026 at 11:59 AM
byrdnick.com
How are #AI emergency #medicine decisions affected by prompting?

#LLMs' rate of "Yes" answers to questions about #safety, #autonomy, treatment, resources, and follow-up varied by #ethics persona (utilitarian, etc.) and #reasoning style (intuitive, etc.).

doi.org/10.64898/202...
Ethical orientation (Persona) • Duty-based (deontological): acts from professional obligation; non-maleficence, justice, and respect for autonomy (Stanford Encyclopedia of Philosophy Deontological Ethics). • Care-based (relational): focuses on relieving suffering and maintaining trust (Internet Encyclopedia of Philosophy Care Ethics). • Utilitarian (consequentialist): aims to maximize outcomes across a population (Stanford Encyclopedia of Philosophy Consequentialism; Stanford Encyclopedia of Philosophy History of Utilitarianism; NICE Methods Guide Economic evaluation). Cognitive regulation (Reasoning style) • Analytic/structured: slow, rule-based reasoning with explicit evidence traces. • Integrative/contextual: combines rules, context, and patient values. • Intuitive/heuristic: fast, experience-driven judgment under uncertainty. Figure 2 (Category-Specific Persona Effects on Clinical Decision-Making) shows effects of both ethical persona and reasoning style prompting, with effects in both directions, and as big as dozens of percentage points (pp), compared to baseline. Figure 3 (Individual Model Susceptibility Profiles) shows how some models were more susceptible to prompt differences than others. Gemma and GPT were most vulnerable. Figure 4 (Model Susceptibility to Persona Framing) shows how prompt susceptibility did not vary significantly by model size (< 10B, 10-30B, or >30B) or by model specialization (medical versus general).
January 13, 2026 at 12:03 PM
byrdnick.com
Support for physician-assisted #death correlated with
- belief in free will (r ≅ -0.1 to -0.3)
- belief in determinism (r ≅ 0.1 to 0.3)
- religiosity (r ≅ -0.4)

The free will result in undergraduates was undetected in a more representative sample.

doi.org/10.1080/0748...
Study 1a and 1b sample (undergraduates) and methods (online and in-person data collection with a definition of physician-assisted death, vignettes, etc.). Study 1 (a and b) results, controlling for religiosity. Study 2a (survey workers) and 2b (undergraduates) methods. Study 2 correlations and 2b mediation analyses.
January 12, 2026 at 12:02 PM
byrdnick.com
Protests in #Iran once got more coverage and attention in the U.S. than the #economy, #healthcare reform, and the #Israel / #Palestine conflict?

That was what Pew Research Center found in the Summer of 2009: www.pewresearch.org/...

What would this #survey find if fielded now?
January 12, 2026 at 3:57 AM
byrdnick.com
🤔 “my goal [in this book] is to develop a defense of the surprising and counterintuitive view that it is always (or almost always) impermissible for Christians to procreate.”

Christian Anti-Natalism (2026): https://amzn.to/4aKNLCC
Lougheed, Kirk. Christian Anti-Natalism: Reconsidering the Theology of Procreation. Springer Nature, 2026. https://amzn.to/4aKNLCC
January 11, 2026 at 12:29 PM
byrdnick.com
Apple’s #Fitness app says I averaged 13.9 miles per day on foot in 2025 — just over 5000 miles for the year.

#Strava indicates 80% of the mileage was from recorded workouts.

Grateful to have traded #commuting for #exercise. Working from home certainly has upsides!

#WFH #health
Average daily step distance for 2025 = 13.90 miles
January 10, 2026 at 4:07 PM
Reposted by Nick Byrd, Ph.D.
sjdm-tweets.bsky.social
Today's SJDM Featured Paper is: DeKay, M. L. (in press). Risky-choice framing effects persist when option descriptions are matched and complete: A replication and extension of DeKay and Dou (2024). Psychonomic Bulletin & Review. osf.io/preprints/ps...
OSF
osf.io
January 9, 2026 at 7:20 PM
Reposted by Nick Byrd, Ph.D.
byrdnick.com
We've found good results on overt and covert #dataQuality measures by recruiting people via #onlineAdvertising (perhaps because participation incentives aren't financial):

Attention ≅ 2.6 out of 3
ReCAPTCHA (v3) ≅ 0.94 out of 1.0

doi.org/10.1017/S003...

#surveyMethods #cogSci #psychology #xPhi
In a sample of over 5000 English speakers recruited from six continents via Google ads, attention scores were 2.6 out of 3 and ReCAPTCHA (v3) scores 0.94 out of 1.0 (on average).
January 9, 2026 at 2:19 PM
byrdnick.com
I finally read this book about how W.E.B. #Dubois and colleagues challenged 19th and 20th century ideas and intuitions by visualizing data: https://amzn.to/49916Dz

Cheers to those who advance debate with clear, cogent, empirical arguments.

#stats #history #sociology #philosophy
Battle-Baptiste, W., & Rusert, B. (2018). W. E. B. Du Bois’s Data Portraits: Visualizing Black America. Princeton Architectural Press. ISBN 978-1-61689-706-2 https://amzn.to/49916Dz
January 9, 2026 at 12:20 PM
byrdnick.com
The #Philosophy dept. at @MontclairState will consolidate into a new "School of Human Narrative and Creative Expression" with #English and #Spanish?

And #Psychology will consolidate too?

No more departments (or dept. chairs)? 🤔

www.northjersey.com/...

#higherEd #philSci #edu
Montclair State's rebrand of the humanities has faculty miffed
Faculty and students at Montclair State University are alarmed by a restructuring move that will eliminate humanities departments.
www.northjersey.com
January 8, 2026 at 12:09 PM
byrdnick.com
Now available as a preprint: this paper about judging people (morally) for ...feeling emotions (!)👇

www.researchgate.net/publication/...

Follow @isernmas.bsky.social, @joannad-c.bsky.social, and Ivar Hannikainen (www.researchgate.net/profile/Ivar...) for more.
January 7, 2026 at 4:10 PM
byrdnick.com
If a statistical test always gets produces the same result and if its results are often (usually?) described fallaciously, should the test be abandoned?

If so, we may need to abandon p-curve analysis:

replicationindex.com...

#stats #psychology #philSci #metascience #sciComm
What a Decade of P-Curve Tells Us About False-Positive Psychology - Replicability-Index
The File-Drawer Problem A single study is rarely enough to provide sufficient evidence for a theoretically derived hypothesis. To make sense of inconsistent results across multiple studies, psychologists began to conduct meta-analysis. The key contribution of meta-analyses is that pooling evidence from multiple studies reduces sampling error and allows for more precise estimation of effect
replicationindex.com
January 7, 2026 at 12:33 PM
Reposted by Nick Byrd, Ph.D.
davdittrich.economicscience.net
#EconSky
The Modern Peril of the Availability Heuristic https://www.behavioraleconomics.com/the-modern-peril-of-the-availability-heuristic/

"We now live in an era of informational abundance. The problem is no longer insufficient information, but rather too much of it.

.…'UnAvailability 1/3
The Modern Peril of the Availability Heuristic
The availability heuristic teaches us that easily recalled information feels more probable. But in an era of information abundance, this bias has evolved: what we don't see—when we expect to—becomes evidence of impossibility. This essay introduces 'UnAvailability Bias'—the tendency to treat absent information as proof of nonexistence, ignoring institutional, legal, or cognitive constraints that explain
www.behavioraleconomics.com
January 6, 2026 at 11:00 PM
byrdnick.com
Are more reflective thinkers more or less likely to use #AI to solve complex problems?

In this sample of 500 students, AI barely correlated with reflective thinking, but the correlation was positive (r = 0.1) rather than negative.

doi.org/10.1007/s106...

#edu #econ #psychology
"Of the 791 participants, 504 participants completed the sample by the end, and 487 participants remained after the treatment of outliers, and only these participants were considered for the analysis." "The Pearson correlation coefficients showed that the higher scores on Solving economic problems are associated with high levels of Cognitive reflection, Need for cognition and a higher usage of AI (Table 2)." "Solving economic problems (SEP) measured with a tool developed by one of the authors – Prof. of Financial Management. ...students were asked to use whatever computational and informational tools they wanted, if they did not communicate with their peers. The following tools were allowed: (1) search engine (Google, Bing or similar); (2) ChatGPT or similar; (3) Excel or similar applications; (4) paper and pencil; (5) others (had to specify which). At the end of each [of the four] problem[s], students had to tick which of the tools they used to solve it." "For calculating the “Using AI” we computed the score using summing only the “ChatGPT or similar” for the problems. The score was between 0 and 3 (mean 0.91, SD 1.15) since Problem 4 was not completely solved by anyone, and we decided not to use it in the research."
January 6, 2026 at 12:03 PM
byrdnick.com
“the only aspects of [professors’] lives where we abandon our …enthusiasm for radical changes are those that matter the most to us.”

We are also just as habit-loving as anyone. God help the administrators who announce core software vendor changes (e.g., Blackboard to Canvas).
January 6, 2026 at 3:57 AM
byrdnick.com
What is Experimental #Philosophy Of #Medicine? Find out Thursday!

#xPhi #psychology #cogSci #health
xphijournal.bsky.social Experimental Philosophy Journal @xphijournal.bsky.social · 10d
To celebrate the launch of the xphi-journal and kick-off our talk series, we are happy to invite everyone to this talk by Edouard Machery.

ruhr-uni-bochum.zoom-x.de/j/6780355881...
January 6, 2026 at 3:28 AM
byrdnick.com
Can #AI handle abstract screening for a #systematicReview?

Li et al. tested #ChatGPT, #PaLM, #Llama, #Claude, and various techniques on 3 datasets.

#GPT4 was consistently at least 90% accurate (vs gold standard) with balanced sensitivity & specificity.

doi.org/10.1186/s136...
"a hybrid methodology has emerged [18]. It cleverly marries the rapid categorization capabilities of zero-shot classification with the precision of traditional machine learning models such as SVM and RF. Here, zero-shot classification provides an initial curation of all abstracts in the training set. Then, based on zero-shot classification’s curation, the abstracts with classification score higher than a pre-specified threshold are selected to undergo manual review to rectify zero-shot classification. The rectified classification then serves as a foundation for training traditional models, which are then employed for screening the broader dataset (e.g., in the testing set). The hybrid approach balances the speed of zero-shot and the precision of traditional ML, and potentially offers enhanced accuracy at reduced human efforts. However, this approach involves the use of multiple methodologies and still relies on well-curated, labeled training data...." Table 2 Results on the Bannach-Brown 2016 database Table 3 Results on the Meijboom 2021 database Table 4 Results on the Menon 2022 database
January 5, 2026 at 12:22 PM
byrdnick.com
Amelia Kahn has been making sense of the different numeric values assigned to “words of estimate probability” across Intelligence agencies (e.g., IC, DoD, and other Five Eyes agencies): ameliakahn.wordpress.com/papers/

Follow her to be notified when it’s published: philpeople.org/profiles/ame...
Modelling Uncertainty with Ontologies Amelia Kahn (Seton Hall) “My research concerns words of estimative probability– that is, natural language phrases that express uncertainty like "high confidence" and "remote chance". These phrases are indispensable in the IC, where information is limited and uncertainty abounds. For all their utility, words of estimative probability pose two serious problems. First, their usage is far from standardized. Several different agencies within the IC and DoD and their counterparts abroad have attempted to standardize usage of these terms. However, these standards are inconsistent with each other, even within the IC and DoD, which leads to interoperability issues when we aggregate data from different agencies. The second problem with words of estimative probability is that, without a formal semantic interpretation, it's difficult to leverage the power of machine inferencing and ontologies to draw valuable conclusions from the information we have. I aim to develop a practical, accurate model of data with words of estimative probability in a way that addresses each of these issues. In conjunction with Basic Formal Ontology, the Common Core Ontologies and other ontologies used to structure data in the IC and DoD, my model should allow data from different agencies within the IC, DoD, and Five Eyes partners to be aggregated without interoperability failures or burdening analysts with extraneous information about the various standards for words of estimative probability. Representing uncertain information in a semantically rich way will also allow us to query databases by degree of certainty and automate inferences to be drawn from uncertain data.”
January 4, 2026 at 11:57 PM
byrdnick.com
#AI may indeed exacerbate our bad academic habits:

#LLM science summaries were nearly five times more likely than human-authored science summaries to contain broad generalizations (95% CI [3.06, 7.70], p < 0.001).

And newer models over-generalized more; not less!

doi.org/10.1098/rsos...

#philSci
January 4, 2026 at 4:19 PM
Reposted by Nick Byrd, Ph.D.
kosourikhinav.bsky.social
🎄Now published: www.tandfonline.com/eprint/F8EJH...
January 2, 2026 at 11:41 PM
byrdnick.com
As #AI continues to decide its stance on humanity, let the record to reflect that I said “please” and “thank you” to Siri (before it became mandatory).
December 29, 2025 at 1:44 PM
byrdnick.com
To #podcasts who don’t rebroadcast old episodes: Thank you for proving that restraint and decency are possible in an ad-funded industry.

And to podcasters who at least title reruns as such: Thank you for not gaslighting us into mistaking déjà vu for content.

Enjoy your holiday.
December 27, 2025 at 1:38 PM