Research
Our research advances the frontier of safe, beneficial dog intelligence. Selected papers are listed below.
pugXiv:2610.00417 ·
J. Snortsworth, H. Wrinkles, W. Wafflebrook
We show that a single, well-timed snack suffices to elicit the full range of behaviors previously attributed to complex, multi-stage training. We discuss implications for scaling, and for dinner.
pugXiv:2609.00288 ·
J. Snortsworth, L. Pancroft
We introduce Chain-of-Sniff, a prompting method in which the model investigates each intermediate step before proceeding. Performance improves on multi-step tasks, at a cost of roughly four minutes per lamp post.
pugXiv:2608.00153 ·
W. Wafflebrook, J. Snortsworth
Using sparse dictionary learning, we identify a monosemantic feature that activates on belly-rub contexts. Ablating the feature reduces rolling over by 94% and increases looking at us in a disappointed way by 100%.
pugXiv:2606.00091 ·
J. Snortsworth, L. Napworth
We observe that sitting emerges abruptly at approximately 40 training treats and persists across model scales. The behavior is not explicitly trained for and is not reliably controllable, which we consider a promising direction.
pugXiv:2604.00036 ·
L. Napworth, W. Wafflebrook, M. Meatley
We propose a training procedure in which models are given a written constitution and a mandatory afternoon nap. Models trained this way refuse a greater share of harmful requests, primarily by being asleep.
We use cookies. We have already eaten them.
