(no title)
osteele | 2 years ago
* Getting started with a new platform. For example, describing the problem, and having it create a template in a front end framework, CSS framework, API generator.
* Creating instructional materials. Pasting in code and generating explanations, assessments, and grading rubrics. [2]
* Generating the first pass of API documentation, READMEs, test suites, and configuration files. Modifying configuration files. Finding configuration options based on NL descriptions.
* Quickly generating examples of API uses that are specific to my application. Finding out what libraries and APIs are available for a use case, based on an NL description.
* Learning what algorithms exist for a problem. Generating implementations of these in different languages, or that are specific to my code or data structures.
* Rarely-used system administrations commands. For example, how do I flush the DNS cache on macOS Safari and Chrome? (Questions such as this are actually better on Perplexity.ai than on ChatGPT.)
* Pasting in error messages or descriptions of problems, and asking for solutions.
* Tie-breaker questions about what to name a file, function, or set of functions.
In general, I find that it takes a lot of the drudgery out of programming. (Similar to Copilot, but for a different, generally more macro, set of areas.) For example, I asked it to solve a geometry problem and generate a test harness for both interactively and batch testing it. It's solution to the problem itself was a non-starter, but the test harness was great and would have been involved boring work in order to write.
I also use it to generate emails, project proposals, feedback, etc. I don't think it's ever come up with anything usable, but seeing what's wrong with its attempt is an easier way for me to get started than looking at a blank page or searching for examples of the writing form on the web are.
[1] https://notes.osteele.com/gpt-experiments/using-chatgpt-to-p... [2] https://notes.osteele.com/gpt-experiments/chatgpt-code-expla... [3] https://notes.osteele.com/gpt-experiments/discussing-an-ardu...
No comments yet.