In one randomized working-paper experiment, conceptual-map scaffolding gave LLM-using novices marginally higher scores (p < .10) and lower score variance, with a marginal benefit concentrated among low-AIC novices¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q3
Currency · newest study 2026 · checked against its sources 2026-10-22 (automated, 2 of 2 entries) · not reviewed by a person

Subclaims¶

q3 i? Scaffolded novices scored higher than unguided novices (M = .45 vs .38), a difference significant only at p < .10. → Bharat Anand 2026 q3 i? The scaffolding × AIC interaction was negative and weakly significant, so scaffolding benefited low-AIC participants most and compressed heterogeneity. → Bharat Anand 2026 (2)

Evidence¶

Bharat Anand 2026¶

Bharat Anand, Lihi Idan. (2026). Generative AI and the Productivity Divide: Human–AI Complementarities in Education and Knowledge Work (Working Paper). https://arxiv.org/abs/2605.18143 doi:10.48550/arxiv.2605.18143

q3 · i? · causal · r2

Scaffolding sub-condition within the randomized experiment's LLM arm: novices given a conceptual roadmap with recommended sequencing versus unguided novices. Variance in post-intervention scores was lower among scaffolded novices (SD=.14) than unguided novices (SD=.22).

"The average post-intervention exam score among scaffolded novices wasM=.45,SD=.14, compared toM=.38,SD= .22 in the baseline LLM group, with a weakly significant difference (p>.05,p<.10)."

Bharat Anand 2026 (2)¶

Bharat Anand, Lihi Idan. (2026). Generative AI and the Productivity Divide: Human–AI Complementarities in Education and Knowledge Work (Working Paper). https://arxiv.org/abs/2605.18143

q3 · i? · causal · r2

OLS regression comparing scaffolded novices to baseline LLM novices with a treatment-type × AIC interaction; the authors interpret scaffolding as compressing heterogeneity by disproportionately aiding weaker extractors.

"The coefficient on the interaction term was negative and weakly significant (p > .05,p < .10), indicating that the marginal benefit of scaffolding diminished as AIC increased."

Discussion¶

Related Claims¶

• Scaffolding improves learning — a broader claim this one bears on
• Minimal guidance is less effective for novices than explicit instruction — a broader claim this one bears on
• AI Interaction Competence, not GPA or prior knowledge, is the decisive moderator of LLM-assisted learning gains — reports the opposite
• Concept mapping improves learning over most comparison conditions by a moderate average amount, but not over time-matched retrieval practice — related
• Extending mandated daily study time from three to four hours produces no significant performance gains for novices — related
• LLM-assisted self-study yields higher post-intervention exam performance than traditional resources among early-career knowledge workers — related
• Without explicit guidance and scaffolding, unguided GenAI use may disproportionately benefit already advantaged students and widen performance gaps — related
• Baseline capability moderates AI productivity benefits non-monotonically across AI's capability frontier — related
• High teaching guidance in concept mapping yields larger effects than low guidance, though marginal after excluding influential studies — related