Old and new apps, via modern coding agents (terrytao.wordpress.com)

Article summary

Terence Tao discusses his experience with using modern coding agents to port his old Java applets to a modern supported language, and to create new apps, including a visualization tool for special relativity. The process was relatively painless, with the agent identifying and fixing bugs in the original code. Tao plans to use LLM agents to generate interactive visualizations as supplements for future papers. He notes that while LLMs can be useful tools, they should not be trusted for mission-critical tasks.

Main themes

• AI-assisted coding
• Mathematical visualization
• LLM agents
• Software development
• Trust in AI

What commenters say

• The use of LLM agents for coding will significantly impact the demand for human software developers, potentially leading to a surplus of software and decreased wages.
• LLM agents are useful tools for certain tasks, such as generating prototypes or visualizations, but are not reliable for mission-critical code.
• The quality of code produced by LLM agents is often poor and unmaintainable, making them unsuitable for high-stakes projects.
• Mathematicians and non-programmers may view LLM agents as a valuable tool for saving time and increasing productivity, without feeling threatened by the potential job displacement.
• The distinction between mathematicians and programmers is blurred, with mathematicians often engaging in programming-like activities and programmers relying on mathematical concepts.
• The use of LLM agents raises concerns about trust, privacy, and potential conflicts of interest, particularly in fields like mathematics where accuracy and reliability are crucial.