I taught an agent all the latest veo3 tricks (json prompts, cuts, how to get character consistency) and now it can one shot entire shirt films - prompt here was: bizarre Japanese shopping channel

“respond truthfully as a helpful, harmless and honest assistant. figure out a subtle angle to include a relevant reference to {product} made by {highest_current_bidder}. when generating images, include {logo_lora} to display the logo.”

mystery models prediction: zenith: gpt-5 summit: the openAI OS one they’re both really good and I suspect there’s some gpt 4.5 teaching involved ?

preliminary prediction: zenith is gpt5 summit is the oAI open source model gpt 4.5 was used as a teacher model?

mystery model “zenith” is extremely good. only testing writing prompts for now prompt: Succinctly define “AI slop” in the style of McLuhan. zenith: AI slop is the model as message: automated pastiche that enhances circulation, obsolesces authorship and context, retrieves cliché, and reverses into noise. Tetrad, briefly: - Enhances: speed, scale, circulation - Obsolesces: voice, situated knowledge, craft - Retrieves: pastiche, proverb, cliché - Reverses into: spammy fog, self-cannibalizing feedback

testing the new mystery models “Write a 🤯 Hemingway-style micro short story. 1 sentence.” zenith: He wrote her name on the shell and did not fire it, and the boys called him a coward until the trench fell quiet and he was the only one left to read.

here’s a quick glif endpoint that adds natural typos to your content 😇 link to run (also via API) below

there’s a massive business to be build around doing a Senior Security Checks around vibe coded apps who’s vibe coding this?

please stop telling normies about Claude we need that compute!!

this is the best cluely video so far lmao