Explore outputs
Every output from the sweeps: pick a model, a vector and a prompt, then drag the strength. The two prompted versions show what the vector's own system prompts produce, for comparison.
Every strength for this prompt
Strength is in Modulo units: 15 is a push as large as a typical token's activations. NLL is how unusual the text is for the unsteered model; ⟳ marks repetition loops. Outputs stop at 80 tokens. See methods.