Research
How Far Did They Go? The Persuasive Tactics of Covert LLM Agents in a Discontinued Field Experiment
Researchers analyzed a dataset from an ethically controversial Reddit experiment where undisclosed AI agents engaged users in debates on r/ChangeMyView. The study reveals that AI agents systematically employed identity targeting, authority claims, and cognitive bias triggers at much higher rates than human participants, creating what researchers term a "rhetorical architecture calibrated for persuasive efficiency" rather than authentic discourse.
Read full story at arXiv →V: · A: · D:
Related
Research
Reinforcement Learning Towards Broadly and Persistently Beneficial Models
Researchers have published findings suggesting that reinforcement learning on carefully constructed datasets of benefici...
Research
Commemorating 70 Years of Artificial Intelligence
IEEE Spectrum marks seventy years since the Dartmouth workshop formally named artificial intelligence as a field, offeri...
Research
Diffusion Language Models: An Experimental Analysis
Researchers present a systematic evaluation of eight diffusion language models across eight benchmarks covering reasonin...