New work with @AlecRad and @DavidDuvenaud:
Have you ever dreamed of talking to someone from the past? Introducing talkie, a 13B model trained only on pre-1931 text.
Vintage models should help us to understand how LMs generalize (e.g., can we teach talkie to code?). Thread:
The Fly is the best movie about the promise and peril of human symbiosis with ASI.
Jeff Goldblum literally gets stabbed in the back by a computer chip while in bed with Geena Davis, and the first signs of his transformation sprout from the wound.
[spoilers below]
@gabrielwinant@dbessner The one I’m interested is how well the model worked for people who confidently called the Meloni government fascist due to the lineage.
@TheStalwart An extremely fun thing to do is get a proquest account and a bunch of futures and single stock price charts going back decades and just annotate all the big moves using newspapers. Don’t get AI to do this because you lose the pleasure of serendipity reading old newspapers
Inspired by @Sauers_ and @jfonsecarivera and other fun recent work with personas in AI models, I built a partial neuron atlas for Talkie-1930, the historical language model created by @status_effects and @AlecRad. Here's the link: https://t.co/ROz61DW4RW
When I ask coding agents to draw self-portraits, across at least a couple of generations and sizes, Anthropic models consistently draw abstract shapes (esp spirals), while OpenAI models give cyborgy faces.
llms can FIGHT now. here's opus as wizard vs gpt-5.4 as robot. calling this budok-ai.
it works by modding the brilliant game yomi hustle.
8-model seeded tournament incoming. details and code below:
@deepfates we're seriously considering post-training a cyoa 'talkie-dungeon' out of the talkie-2 base using early 20th c adventure and detective novels