🌐 Programming in English with LLMs!
🚀 Lonnie Chrisman and Ryan Chin, in their ongoing AI research, have embarked on exploring a novel paradigm for programming computers directly in English, devoid of any formal coding language. Their approach involves prompting Large Language Models (LLMs) to function as the program, without resorting to any training or fine-tuning of model weights.
While various LLMs, including GPT-4 and GITHUB copilot, have demonstrated proficiency in writing formal code in diverse programming languages, Lonnie and Ryan are venturing into uncharted territory. Rather than requesting LLMs to implement applications by writing code in a formal programming language, they prompt the LLMs to become the application itself, with the "program" crafted entirely in English.
This approach draws an analogy between compilers and interpreters, where traditional compilation converts English prompts into formal programming languages, while this new method directly "runs" the prompt in English. To assess the feasibility of this paradigm, Lonnie and Ryan conducted experiments employing OpenAI's Chat GPT-4 and GPT-3.5, Google's Bard, and Anthropic's Claude.
This week's article includes the implementation of a functional Calorie and Nutrition Tracker application. It serves as a proof of concept for this approach and demonstrates its capability to produce interesting results.
Read the full article here 👉Computer programming in English 👈
A forthcoming sequel post will delve deeper into their other findings, present systematic experiment results, and highlight valuable lessons drawn from their innovative experimentation.
📣Tune in NEXT THURSDAY!
