English is the hottest programming language // any natural language actually

Here is a summary of the key points from the transcript:

- The speaker, Dr. Matt Welsh, argues that 50 years of research into programming languages has failed to solve the fundamental problem that programs need to be written, maintained, and understood by humans. Humans are not very good at these tasks.

- Today, AI models like GPT-3 allow instructions to be provided in natural language without needing an explicit program. Welsh believes this will eventually replace programming entirely as we instruct language models directly.

- Welsh discusses the history of programming languages and AI capabilities, noting the rapid recent advances. He argues the dialogue has shifted from “computers can’t do that” to “AI might destroy society.”

- He predicts the future evolution of programming will be humans instructing language models, without traditional programs. The “model is the computer.”

- Welsh demonstrates examples of instructing GPT-3 to perform tasks like summarizing a transcript. He notes you must use certain phrases like “do not” in all caps to properly instruct the model.

- He argues AI code generation will replace most programmers. The cost to replace a developer with GPT-3 is about $0.12 per day.

- Welsh discusses how software development may evolve, with product managers specifying requirements to AI systems. Testing will remain important.

- He demonstrates his startup Anthropic’s tech for building customized chatbots with natural language instructions. He argues new abstractions are needed.

- Welsh concludes computer science education must evolve, but foundational knowledge will remain important. There are still many open questions regarding limits and validation of language models.

// in progress…