The other day I remembered the concept of AI winter and realized we are now probably living through an AI summer.

Against this backdrop, and using LLMs every day at work, I want to share three quick thoughts.

1. UI through AI: at work, we are increasingly relying on textual interfaces using natural language to express complex business logic. The natural language of the rule is then translated to a deterministic query language by a LLM. If we didn’t have an LLM to do this for us, the alternative would be a cumbersome UI full of dropdowns or connectable diagrams. Already, then, it is possible for AIs to do this mapping between user desires and business logic that backends understand. AI starts to be a core part of the UI.
2. Is in-house training becoming obsolete? Also at work, I’m working on intelligent document processing. The combination of deterministic OCR and a LLM that interprets the result is starting to give excellent results, becoming comparable to IDP solutions that are trained on a lot of data. Does this mean that the LLM models are now so powerful that the traditional training of models starts to shrink? Also recently, I read the Replit CEO talking about how a new product they launched, Agent, didn’t rely on their own data for training and still achieved great results just using Anthropic’s Claude 3.5 for generation of apps with LLMs.
3. Rising temperature: LLMs have a temperature parameter that allows them to be either more precise or more creative. I personally feel that the LLM wave is raising my own personal temperature. I’m making bolder bets, in what and how I build. I’m moving faster. I’m more energetic and more anxious about the future. It feels like a bit of a fever.