OpenAI is lying

Obsidian has beautiful UI

I want to fine-tune a model so it can create websites like year 2000 geocities.

So much drama AI scene becoming a season of Gilmore girls.

All models have the "pink elephant" problem where design guidelines leak into the text content of the output. But Gemini 3.1 pro is still a cut above the rest when it comes to creativity. But Google, for the love of god, just cannot RL/fine-tune it for agentic work and their harness engineering is somehow even worse. It's like working with a gifted disabled artist where as Opus and GPT are multi-armed productivity monsters who are near-blind.

UPDATE: Remember that Uncodexify project I shared in my last vid about OpenAI and frontend? Turns out OpenAI noticed it and asked about the license just a few days before posting this article lol https://github.com/cyxzdev/Uncodixfy/issues/5

“You’re holding it wrong” energy

It’s so funny how these models think “no cards” means use a card and just get rid of the rounded/squared outline and change nothing else

Feels like the HTML of the regular front-end I have to work.

About your last take: ask GPT and Claude what is the last version of Vulkan (tell them they can't search for it) - GPT says 1.3 - Claude says 1.4

"What the slop is going on" could really catch on as a phrase