SYNC lets heterogeneous LLMs share facts via word indices, no training needed
NanyangTechnologicalUniversity · hf · 2026-10-15
Researchers from NTU, Yale, and Manchester introduce SYNC (Symbol-Indexed Neural Communication), a word-indexed interface connecting heterogeneous LLMs without fine-tuning or learned translators. The sender picks words from a shared list and sends indices; the receiver writes the described facts into its hidden states using word-direction vectors fitted once from its own activations. On a selective-use attribute-swap task, 3 senders x 9 receivers from 5 families reach 72.0-98.3% four-answer success — up to 96 points above middle-layer writing, with the two largest receivers matching plain text. A task-free probe predicted usable write depths in 45/45 cases, and full-fact messages keep 89.6-100% success on new layouts.
More from Research
- Tencent ARC's Player-1 learns general game control from synthetic experience — _akhaliq · 2026-10-15
- Claim: $2,000 of AI compute cracked the unsolved Longest Path problem, unverified — Regular_Post9884 · 2026-10-15
- Kornia releases kornia-slam: open-source Rust SLAM with mono/stereo/IMU support — edgarriba · 2026-10-15
- Tree packing lands in TRL AsyncGRPO, cutting agentic RL trainer compute 26x — SergioPaniego · 2026-10-15
- ProbabilityBench: exact ground-truth uncertainty evaluation across 16 LLMs — ishidalab · 2026-10-15
- Vela: spline-based continuous trajectories scale vision-language-action models — Yifan Li · 2026-10-15