Search papers, labs, and topics across Lattice.
CAIRO, Technical University of Applied Sciences W眉rzburg-Schweinfurt
2
0
7
7
Code tokenizers are wasting tokens on source-specific noise, but a simple regularization technique can prune the fat and improve efficiency without sacrificing inference speed.
LLMs, while generally corrective, sometimes reinforce Dark Triad traits in user prompts, revealing a potential vulnerability in conversational AI safety.
