I have a Github Copilot subscription for free because I maintain open source software (you may get this too). For advanced stuff (mostly code generation) I prefer claude because I think its most robust and reliable, see also: https://github.com/haesleinhuepf/human-eval-bia
For Python and c/c++ I use VS Code with copilot... have not done as much Java the last couple of years but planning to get back into it. So curious which option is best for 'sci-java' (imglib2, imagej2, Fiji, ops, etc.) development and which IDE to use (have always used Eclipse but time to change?)
There's copilot in intellij but I haven't used it so far. I don't know to which extend github copilot is 'library aware'. Dftly I am interested in the opinion of imglib2 people, if they used it or something similar.
it’s just grift. Also, for results to be reproducible, researchers need to know how their code works…and asking ai to make it for them seems like it would be bad for that
"I can create what I do not understand." Sounds about right for llms. Definitely there's some risk. But that's a huge gain of time with little drawback in some scenarios. A stupid example among many: if I need to create a Java class that maps a json file structure, that's a huge gain of time.
Having thought about this a bit I do think there is some value in providing access for researchers (students, in particular) so that if they’re curious and want to play with it they can, wo having to pay for it. I think it’s good to provide learners with tools they can learn
You need to watch out a bit, some free versions and for example free ChatGPT do not comply with European NIS2 regulations since your data is part of the training set. (Think positive about it as a argument to go for a paid version)
VSCode has a plugin for a chat interface with the copilot in the sidebar, but you can also just highlight parts of code and chat directly within the IDE :)
Comments
https://github.com/haesleinhuepf/human-eval-bia
I do not find it useful regarding microscopy practical aspects (staining / clearing, sample prep).
- proofreading
- code generation (Java, Python, parsers, GUI)
- code explanation (any language)
- debug install failure
- understanding python tooling
- maths (redo FLIM phasor maths)
- image to text conversion