3 ms·
When I write code it's usually not working the way I want it on the first try. There are errors and it also doesn't do exactly what I and users would want it to
by galaxyLogic 3y ago
When I write code it's usually not working the way I want it on the first try. There are errors and it also doesn't do exactly what I and users would want it to do. So I iterate. I compile and test and debug and modify.
But how could an LLM do that kind of thing if it can not actually run the code and inspect the results, and compare them to some criteria it has in its "head"? Does any LLM do such a thing? Of yes you say surely next year but let's focus on what we know to be possible now.
- CharlesW 3y ago> But how could an LLM do that kind of thing if it can not actually run the code and inspect the results… https://openai.com/blog/chatgpt-plugins#code-interpreter https://openai.com/blog/chatgpt-plugins#code-interpreter
- williamcotton 3y agoI’ve been working on a tool called autorails. It is a GitHub application that reads issues and create PRs. The ReAct tool in LangChain with a custom DockerGitGitHub tool results in this PR: https://github.com/williamcotton/autorails-alpha/pull/38 https://github.com/williamcotton/autorails-alpha/pull/38 The log of the work done by the LLM is in the PR description.