Post-training is what gives LLMs a distinct voice. As long as post-training exists, their outputs will remain detectable.
New Substack post, written by our own @bradley_emi. Link below
We benchmarked on PERSUADE, a collection of nearly 26,000 student essays each associated with diverse author demographic information like age, gender, and disability status. Pangram 4 demonstrated zero false positives across all examples.
Link to the full breakdown below
Can you tell AI-generated drama from the real thing?
Pangram is running an in-person event in NYC! Participants will get on stage and recite either a true personal story or a random AI-generated story. Audiences will vote to decide what's real.
Signup below!
