Fetching from the wire…
01
02
03
04
05
06
07
08
Research2026-09-25 · source-backed
arXiv 2609.29045 points out that any input string has many valid tokenizations, and non-canonical ones route around localized edits and unlearning. Across five LLMs, six datasets and six editing/unlearning methods, 38.6% of alternative tokenizations bypassed the modification, using only the released model with no pre-edit weights, training data or classifier. Any claim that an open-weight release "removed" knowledge needs testing beyond the canonical tokenization.
Each link below shares sources, entities, or timing with this story.
It avoids the correlated near-duplicates that similarity-aware and conflict-aware defenses filter on. Across two QA datasets, three LLMs and nine RAG configurations it matches or beats prior attacks in most settings, with its biggest gains against clustering- and conflict-awar...
Hand a model a correct program and tell it to find and fix bugs. It will find bugs. arXiv 2609.10123, posted September 9, ran LLMs as blind iterative bug-fixers across multiple models and repair environments. The headline result is the ratio: the rate at which these loops dama...
Thirteen authors ran a generational genetic algorithm over specialized agents that separately handle mechanistic argument, assumption reconsideration, and evidence and testability assessment (arXiv 2609.15938). Evaluated against DepMap and Open Targets across 34 cancer types,...
First benchmark of off-the-shelf LLMs against expert-derived ground truth built on INCOSE criteria, ten models across two families and five generations each, one hundred independent runs, two requirement sets, five temperatures. The error profile is asymmetric, and performance...
This comparison ran the baseline the retrofitted-linear-attention literature skipped. Across multiple LLMs and downstream tasks SWA with sinks matches or beats post-trained linear attention, and on Needle-in-a-Haystack and BABILong it scores 2 to 10 times higher. The recommend...
A prespecified randomized audit ran seven models over 3,024 choice sets, three personas, nine paraphrases and nine arms for 40,068 scored responses (arXiv 2608.14399). Reputation dominates, with a 3.9 to 4.7 rating raising choice probability 31.4 points. But demographic parity...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.