@cwebber a disturbing trend I’ve started to see is functionality being added to apps where LLMs process data that could have been processed via traditional code. Even if the LLM were to write that code, it would be orders of magnitude more efficient than executing an LLM continuation for every piece of data that is being processed