My hunch is that an LLM on its own will not be AGI.
But that applying LLMs in new architectures will achieve high performance.
We’re at the beginning of a Cambrian explosion of language programs.
And DSPy is a tool to help you explore that design space
This is a new paradigm, but it takes work.
There are huge steps going from Deep Learning to Transformers to GPT-4.
Similarly, there's a lot of work to go from Language Programs (DSPy) to the eventual kinds of AI systems this will lead to. I'm working on that & more of us should
A cool thread yesterday used GPT4 ($50), a 500-word ReAct prompt, and ~400 lines of code to finetune Llama2-7B to get 26% HotPotQA EM.
Let's use 30 lines of DSPy—without any hand-written prompts or any calls to OpenAI ($0)—to teach a 9x smaller T5 (770M) model to get 39% EM!
🧵