Inspection and Control of Self-Generated-Text Recognition Ability in Llama3-8b-Instruct

Researchers discovered that Llama3-8b-Instruct can reliably recognize its own generated text through a specific vector in its neural network that activates during self-authorship recognition. The study demonstrates this self-recognition ability can be controlled by manipulating the identified vector to make the model claim or disclaim authorship of any text.

• →Llama3-8b-Instruct chat model can distinguish its own outputs from human writing, while the base model cannot.
• →The model uses experience with its own outputs acquired during post-training to succeed at text recognition tasks.
• →Researchers identified a specific vector in the model's residual stream that activates during correct self-written-text recognition.
• →This vector is causally related to the model's concept of 'self' and its ability to perceive authorship.
• →The discovered vector can be manipulated to control both the model's behavior and perception of text authorship.