Santiago Valdarrama

Using "a model" is exactly what I'm doing right now: Jev is a model. When you use Anthropic models, OpenAI models, Grok, or any other provider, you are using a 3rd party component. There's nothing different here than using Jev. But focusing on the main claim: Yes, you could "write a custom classifier" with another model, but there are a few issues with this: 1. You will have to do the work. Why would you spend time reinventing the wheel? 2. A custom classifier built with an LLM has a problem: confidence scores! I need those. Jev provides those. Your model doesn't. 3. Jev is trained to return reasonably good confidence scores. These scores are supposed to be calibrated for many different problems (although it's impossible to calibrate for every problem). 4. Jev is way cheaper than using a regular model. https://x.com/jaxcore/status/2113626554144272674?s=20