Finding: Practice-based programming instruction (with elaborated error feedback) outperforms video-based instruction on transfer performance, metacognition (reduced overconfidence), and motivation, suggesting errors are productive learning signals rather than obstacles.
Why it matters: Directly extends the "Less Talk, More Code" finding and has strong implications for learning sciences, skill acquisition, and the design of AI-assisted tutoring systems — all core teaching interests.
Method: Two experiments comparing video vs. practice-based instruction with elaborated feedback, measuring transfer performance, overconfidence, and engagement/interest.
Jargon: *Transfer* = applying learned skills to novel problems; *elaborated feedback* = error feedback that explains why the answer is wrong, not just that it is wrong; *metacognition* = awareness and regulation of one's own learning/performance.
⚠️ No control group mentioned | Single study (not replicated) | Limited statistical detail in abstract
Dr. Ankit Khare, Prakriti Khare, Nisha Siddha • Semantic Scholar
Finding: Dark patterns in Indian e-commerce are systemic, exploiting System-1 (heuristic) cognition to manipulate consumer decisions, with disproportionate impact on low-literacy users; existing regulation is insufficient and reactive rather than preventive.
Why it matters: Directly addresses dark patterns, ethical persuasion, cognitive biases, and choice architecture—core topics in behavior change design and UX ethics curricula.
Method: Qualitative secondary-data analysis drawing on regulatory reports, national audits, consumer surveys, and judicial decisions.
Jargon: *Dark patterns*: deceptive UI/UX designs that steer users toward unintended choices; *Nudge sludge*: friction deliberately added to obstruct beneficial or reflective choices (opposite of a helpful nudge); *System-1 cognition*: fast, automatic, heuristic-driven thinking vs. slow, deliberate System-2 reasoning.
Shirley Shimin Qiu, Mable Zhou, Jonathan Schooler • PsyArXiv
Finding: Multi-agent LLM teams explore more broadly than solo agents but perform *worse* on insight problems, because coordination disrupts convergent thinking—a structural cost that extra time doesn't fix.
Why it matters: Directly relevant to AI impact on collaboration and collective intelligence, and raises important questions about when AI teaming helps vs. hurts—useful for teaching team dynamics and human-AI collaboration design.
Method: 2Ă—2 experiment (solo vs. two-agent Ă— time budget) with multilevel mediation decomposing coordination costs into protocol failure (47%) and residual convergent-integration impairment (53%).
Jargon: *Insight problem solving* = problems requiring a sudden representational shift ("aha moment"), as opposed to incremental analytic problems; *convergent integration* = synthesizing explored information into a single correct answer; *MAS* = multi-agent system.
⚠️ Small sample (n=82) | No control group mentioned | Single study (not replicated)
Y. Xin, Lawrence J. Jin, Jessica Fong et al. (4 authors) • Semantic Scholar
Finding: A neural autopilot model of habit formation, applied to social media posting behavior, shows that forced experimentation alone doesn't create persistent habits—reducing reward volatility combined with forced experimentation does.
Why it matters: Directly relevant to behavior change design, habit formation mechanisms, and persuasive technology/engagement design in social media platforms, with implications for understanding how platforms engineer habitual use.
Method: Structural estimation of a neuroscientifically-grounded habit formation model fit to individual-level posting data from a Chinese social media platform across pre/during/post-COVID lockdown periods.
Jargon: *Neural autopilot model* — a habit model where behaviors become automatic/stimulus-driven rather than goal-directed, grounded in neuroscience of basal ganglia; *state-dependent habit model* — traditional economics model where past consumption directly enters utility; *reward volatility* — variance in the rewards received from a behavior, not just average reward level.
⚠️ No control group mentioned | Single study (not replicated)
Widya Khafifah, M. Hatim, Ida Suryani • Semantic Scholar
Finding: Digital comic learning media significantly improves fourth-grade students' learning outcomes in Pancasila Education compared to conventional lecture-based instruction.
Why it matters: Marginally relevant to learning sciences and educational media design, but the context (Indonesian civic education, elementary school) is too narrow and applied to be useful for the professor's core topics.
Method: Quasi-experimental design with nonequivalent control group, pretest/posttest, independent samples t-test.