UVA talk: Generative Psychometrics uses LLMs to generate and validate test items
GolinoHudson · x · 2026-11-13
Hudson Golino (University of Virginia) gives an online NCME talk on Nov 18, 4pm ET: Generative Psychometrics for Automatic Item Development and In-Silico Structural Validity.
The framework combines LLM item generation with computational structural-validity checks: construct definitions guide generation, LLM embeddings feed Exploratory Graph Analysis, unique variable analysis flags redundant items and bootEGA flags unstable ones — providing in-silico structural evidence before collecting human data. He'll also cover how model dependence, prompting, and construct definitions affect results, positioning in-silico evaluation as a transparent screening and refinement stage for educational assessment development.
More from Research
- Lichess bibliography crosses 300 research papers, with 100 published in 2026 alone — christopher · 2026-11-14
- KosmosCompactBench benchmarks compaction algorithms; Prime Intellect best open source — BenBlaiszik · 2026-11-14
- Stanford COLM 2026 paper: LLM-generated teaching content is far less diverse than human work — HariSubramonyam · 2026-11-14
- Masked decoding trick tames text rivers in Gemma 3 12B, with a width-aware LoRA and coherence RL — cephaloform · 2026-11-14
- Most Models Misbehave Without Explicit CoT Reasoning — ZhijingJin · 2026-11-14
- AM-Bench: 35 Scenarios, 8,750 Trials Expose Frontier Model Blackmail — ZhijingJin · 2026-11-14