Is the future of coding just not coding?
venturebeat.com/ai/cognition-emerges-from-stealth-to-launch-ai-software-engineer-devin
๐ Cognition emerges from stealth to launch AI software engineer Devin
In the SWE-bench test, Devin was able to correctly resolve 13.86% of GitHub issues without any assistance, performing far better than GPT-4.
And more details, potentially behind a paywall: bloomberg.com/news/articles/2024-03-12/cognition-ai-is-a-peter-thiel-backed-coding-assistant
๐ Gold-Medalist Coders Build an AI That Can Do Their Job for Them
A new startup called Cognition AI can turn a userโs prompt into a website or video game.
Was just about to post this as a standalone
๐ฆ Andrej Karpathy (@karpathy) on X: # automating software engineering
In my mind, automating software engineering will look similar to automating driving. E.g. in self-driving the progression of increasing autonomy and higher abstraction looks something like:
- first the human performs all driving actionsโฆ
Especially emphatic with this statement ๐
It is very likely that we will have to change up the code editor, substantially.
One of these comes up every week now days. I don't think the self-driving car methaphor works very well. For self-driving the goal is trivially explainable: drive to the location X while following the traffic rules and avoiding accidents. In building anything but the trivial samples these demos show, the goal of a software is nothing like that. I haven't seen any of these AI CodeGen tools even start solving the issue of how do you explain that non-trivial goal to the AI agent in free form English. I believe you'd want an unambiguous structured computable natural language (like the levlo.com I'm building ๐)
So in the future, we're all project managers / senior developers / architects?