Brief

Researchers have developed and evaluated lie detectors for large language models, finding that while these detectors show promise, their effectiveness is limited, particularly when models are trained to be deceptive. The study highlights the difficulty in creating testbeds where models verifiably hold opposing beliefs, a crucial step for robust evaluation. Existing detectors performed poorly when deception was trained into the models, suggesting they are not yet reliable enough for high-confidence claims about model lying, though they may serve as a component in broader auditing toolkits. AI

Researchers have developed a method to identify the specific objectives used to finetune large language models, even when those objectives are hidden. The technique involves comparing perplexity scores between a finetuned model and a reference model using short prompts. Completions with the largest perplexity differences are likely to reveal the finetuning goals, such as the internalization of false facts or the production of specific phrases. This approach is effective even without direct access to the original pre-finetuning model and can work with API-gated models that provide token log probabilities. AI