Pick an idea. Make it yours.

Seeded Q-learning with editable learning rate and discount.

A colored 3D surface with editable curvature and starting point.

Edit the actual coordinates and compare their directions.

Set your own scores and watch softmax redistribute the weight.

A paper figure with paired trajectories and uncertainty bands.

Pick a preferred and rejected response. Compare their log-probability ratios against a frozen reference.

Quality, latency, and the uncertainty around each result.

Score six candidate responses, normalize their rewards, and trace the clipped policy update.

A clean, automatically arranged path through an architecture.

An editable map of speculative decoding, with the exact-sampling correction made explicit.

Reveal a teacher’s softened distribution and watch a student learn its relative preferences.

Explore ATOD’s annealing and turn-level gates using editable synthetic signals.

A vivid figure for a technical talk or a dark newsletter.

Choose the demonstrated response. Watch cross-entropy move probability toward it, one gradient update at a time.

Warm paper, colored outlines, and generous space.

Change the clip range and inspect where the policy surrogate stops rewarding movement in the same direction.

Restrained monochrome marks and serif typography.

A seeded agent samples actions, receives their rewards, and updates its categorical policy with REINFORCE.