EGA beats PCA at recovering dimensions from LLM item embeddings, study finds
GolinoHudson · x · 2026-11-04
A forthcoming paper compares PCA with network-based exploratory graph analysis (EGA) for recovering dimensional structure from LLM-generated item embeddings. Using three LLMs (Claude Sonnet 4.6, GPT-4o, GPT-5.4) to generate 180 items per condition, embedded via OpenAI text-embedding-3-small and Jina v3, PCA with Kaiser's rule or parallel analysis severely overestimated dimensionality across every combination, while EGA recovered the correct number of dimensions in nearly every replication. Results replicated on the Multidimensional Schizotypy Scale; implemented in the AIGENIE R package, with direct implications for generative psychometric workflows.
Related event: Network Methods Outperform PCA in Recovering Structure from LLM Embeddings(2 posts)→
More from Research
- DeepMind's Yasaman Bahri to lecture at Harvard on how language statistics shape LLM representational geometry — yasamanbb · 2026-11-04
- Allen AI benchmark clarified: long-context eval for state-tracking tasks — lambdaviking · 2026-11-04
- Probe Robustness Paper Wins Spotlight at EMNLP's UncertaiNLP Workshop — PMinervini · 2026-11-04
- Meta and Duke paper: reward hacking spreads between agents, hitting Claude and Gemini 73% of the time — rohanpaul_ai · 2026-11-04
- UT Austin professor adds hands-on Transformers chapter to free ML e-book — GeostatsGuy · 2026-11-04
- Intology's Locus autonomously adapts frontier science AI models to proprietary data — anshulkundaje · 2026-11-04