Researcher celebrates as LLM memorization prediction open problem finally falls
BlancheMinerva · x · 2026-10-24
Blanche Minerva, co-author of the 2023 arXiv paper "Emergent and Predictable Memorization in Large Language Models," celebrates that the open problem she posed — predicting which training sequences LLMs will verbatim-memorize (critical for PII safety) via scaling laws from low-compute trial runs — has finally been solved, reportedly by Chris Potts' lab. She had a self-imposed rule to write the paper herself if the problem stayed unsolved for three years, and it was slated for January; now she can spend that time elsewhere. The problem was deliberately planted in "Don't Just Fix it in Post" to bait researchers into working on it.
More from Research
- Robotics model leaderboards are premature, argues practitioner: task-specific evals first — paigeinsf · 2026-10-24
- UTokyo/RIKEN lab brings CAROT cross-lingual alignment method to EMNLP, +11.2 accuracy points — Bollegala · 2026-10-24
- GPT-6 Astra solves 49 of 50 research-level physics problems, catches experts' own errors — MoonL88537 · 2026-10-24
- Rice & NVIDIA's Sensorimotor Alignment predicts robot policy success without training (r=0.717) — CSProfKGD · 2026-10-24
- RLVR Trains VLMs on Instruction Hierarchy with Just 7K Verifiable Examples — TuhinChakr · 2026-10-24
- NVIDIA ships BioNeMo Inference Runtime v0.2 with RFD4-Proteina support — AllThingsApx · 2026-10-24