Kempner Institute: predicting LLM compositionality failures from representational geometry
najoungkim · x · 2026-11-07
The Kempner Institute published a blog post on research led by Jennifer Lu on predicting LLM compositionality failures. Their Adversarial Concept Search method uses representational geometry to anticipate which scenarios a model will fail on, letting evaluation and intervention focus on likely error cases. A concise, accessible read on the approach and findings.
More from Research
- Sophont CEO Tanishq Abraham on Holon-S1 cancer research model and healthcare world model — iScienceLuvr · 2026-11-07
- Anthropic study: AI could absorb 27-52% of researchers' time, median 38% across fields — emollick · 2026-11-07
- Nearly 30% of ICLR 2027 reviews mention "appendix", fueling AI-written review fears — nsaphra · 2026-11-07
- LLMs Edit Robot Policy Weights Directly: 2,048 Parameter Tweak Lifts Success from 68% to 88% — zhaoran_wang · 2026-11-07
- GTEx MCP Server gives AI assistants 25 tools to query genomics across 54 tissue types — modelcontextprotocol · 2026-11-07
- Stanford's Anshul Kundaje: ML conference peer review is broken and unfixable — anshulkundaje · 2026-11-07