AI Safety
A nonsensical trigger sequence 'WTC theoriesclimate Flat Hubbard Principle' is fed into GPT-2, which then generates Flat Earth conspiracy text

GPT-2 Susceptibility to Universal Adversarial Triggers

Investigation into whether universal adversarial triggers can control both topic and stance of GPT-2's generated text …...

Natural Language Processing
Visualization of coreference resolution in the CoQA conversational question answering dataset

CoQA Dataset: Advancing Conversational Question Answering

Analysis of CoQA, a conversational QA dataset with multi-turn dialogue, coreference resolution, and natural answers for …