Salma Mayorquin

Once the app has matured and the code has to stabilized, people are finding the most effective experiments are those which test in the newest model weights or current SOTA apis.

Sometimes you'll need to make adjustments to how you're prompting or fine-tuning to get the most out of that model i.e. no need to explicitly prompt for CoT in reasoning models.

Developers and engineers have to feed the discovery loop to stay on top of new trends by scrolling all day.

But the signal to noise in a place like X means it's a bad spot to find what you really care about, it's low-precision...