AI Beyond LLMs

Been playing around with Jev from TypeSafe AI recently. For the last couple of years, whenever we say AI, we almost automatically mean ChatGPT, Claude, Gemini and other LLMs. Give them some information, ask a question, and they give you an answer. Incredibly useful.

But while experimenting with Jev, I noticed how often we're asking AI to do something that doesn't actually need an answer in words.

Take an invoice coming into a company. You don't need AI to write out "based on the information available, this invoice appears to be legitimate..." You just need it to tell your system the invoice is 97% valid, 2% duplicate, 1% needs review, and let the software take it from there.

That's the difference with Jev. It flips the current expectations we have with AI. It's designed to make decisions that software can directly use.

There are a lot of places where this could simplify things. In some of my recent products we were relying on LLMs for classification too, like identifying the document type or the validity of an invoice or expense bill. LLMs were getting increasingly capable of handling such tasks, but the Jev approach makes it faster and much cheaper instead of consuming an obscene amount of tokens with the most capable LLMs.

Products now will use a combination of AIs, each bringing its own strengths to deliver a seamless experience to users. LLMs are great when we need language, reasoning, creativity and interaction. But plenty of situations just need a yes or no, a number, a probability, a category, a decision.

Jev feels less like replacing traditional software with AI and more like giving traditional AI a bit of judgement. Sometimes the software just needs to make a better decision, and Jev fits in well there. We need to experiment a bit more with its capabilities, and maybe use it in a working product to know better.