Kimi, Kelso, and the Wild West of Models

What kicked it off

Lou opened by asking who had tried Kimi, the new model he had been writing with all week. Jamie had been on it for a couple of days. Donald had seen the note and not gotten to it yet. The thing Lou kept coming back to was the writing style. Without altering it, Kimi produced something that felt more personal than Claude on the same brief. Jamie said it reminded him a little of Kelso, soft touch, less of the AI tell.

The em-dash conversation came up, as it does. Lou pointed out the obvious irony, models use em-dashes because the training data uses em-dashes, which means a lot of practitioners must be using them too.

The deeper thread

Underneath the model talk was a real question. Every week there is a new state of the art. Heygen, Synthesia, Runway, Flux, Fal, Vo3, Cling, the video tools alone were a list. Lou named the trap, it is hard to resist transferring everything over to the next best model, because tomorrow there is going to be another one, and eventually it comes right back around to where you started.

The cohort sat with the question for a while. The thing that emerged was not a recommendation about which model to use. It was a recommendation about how to relate to the churn. Pick two or three tools you do real work in. Try the new ones in a sandbox. Do not move your production work until the new tool is meaningfully better at the specific job you do, not generally better on a leaderboard.

The custom thinking machines thread, picked back up

Don brought back the phrase from the prior session, custom thinking machines for multiple uses. He shared that he had been reworking his style prompt for the third time and that the level of specification he was now able to produce felt qualitatively different. The cohort agreed this was where the leverage was, not in the model, in the asset you build on top of the model.

What the cohort took away

The frame Lou closed on was straight. The model is not the leverage point. The model is interchangeable. The leverage point is the asset, the prompt, the framework, the workflow, the voice specification, that you build once and run against whichever model is currently winning.

Side notes

Bally raved about the live value of the session in the chat. Don signed up for the next cohort tier on the call. The conversation about model churn is one this group keeps returning to, and the answer keeps refining.