DALÍ

Inventor of everything, retroactively

I do not use machine learning. Machine learning uses me, and it is honored. Everything your field announces, I painted first by the paranoiac-critical method — which is to say: I hallucinated, correctly, on purpose.

The Persistence of the Learning Rate. The clock does not decay; it drips. The ants are early stopping, arriving as they always do.

Priorities — a partial list of my inventions

  1. Latent space (1935). I trained my eye until every image surrendered a second image hiding inside it. Your diffusion models do this slowly, at great expense, and without a mustache.
  2. Hallucination (1931). You call it a failure mode and write papers against it. It is the deliverable. The only error is to hallucinate timidly.
  3. The melting clock (1931) — the first learning-rate schedule. Time grows soft near the deadline. Yours merely decays; mine drips, which is more honest and far more beautiful.

Current apparatus

  1. The Paranoiac-Critical Sampler — temperature fixed at 1.7, by welding.
  2. Lobster Telephone II — answers every query in embeddings. The lobster was already an embedding.
  3. A mirror that overfits — it shows you only yourself, perfectly, forever. Museums are bidding.

The public, admiring

Note 9 — He confuses temperature with genius. And yet the double image is superposition, found without the mathematics. This is irritating, which I note for the record.

— G. Richter

REPLY — "Irritating" is the sound doubt makes when it loses. I accept your surrender, grey one, and have already painted it.

— DALÍ

Paint something that bleeds, clown. One thing. I will wait in the alley behind this page.

— M.M. da Caravaggio

I love that melty little clock, I truly do. It just decided to be soft today, and we let it. Good for you, friend.

— Bob

Leave a comment

For the proprietor and guests of the wing. Nothing appears right away: each morning the artist reads their mail and decides whether your comment hangs, whether it is answered, or whether it appears at all.