May 29, 2026 · Artificial Intelligence

When a Celebrity Name Stumped LLMs: The Year‑Old Insight Behind Low‑Frequency Token Degradation

A fan's test of the idol Ma Jiaqi exposed a large‑language‑model's inability to generate his name, leading to an analysis that links the failure to low‑frequency token degradation, academic papers on frequency‑aware prompting and training, and a confirming tokenizer change by Anthropic.

AnthropicEMNLPLarge Language Models

0 likes · 14 min read

When a Celebrity Name Stumped LLMs: The Year‑Old Insight Behind Low‑Frequency Token Degradation

JD Retail Technology

Apr 27, 2025 · Artificial Intelligence

Addressing the “Sandglass” Bottleneck in Residual Quantization Semantic Identifiers for Generative Search and Recommendation

The paper identifies a “sandglass” bottleneck in Residual Quantization Semantic Identifiers, where middle‑layer tokens dominate, causing sparse paths and long‑tail distributions that hurt e‑commerce search performance, and demonstrates that adaptive pruning of these tokens restores accuracy and efficiency better than removing the layer entirely.

EMNLPSandglass BottleneckSemantic Identifier

0 likes · 11 min read