It looks like there is a lot of interest in the Gr...

# gsoc

[James Ward]

It looks like there is a lot of interest in the Groovy -> Kotlin Gradle converter project! I shared this in a thread but figure it is useful for many here:

My own thoughts on the Gradle DSL converter are that it'd be cool to see if you can leverage OpenAI or some LLM with a corpus of GitHub projects that have done a build.gradle to build.gradle.kts migration.

[Priyank Shankar]

Maybe it's my bad, but I am unable to locate the thread

[James Ward]

Someone had asked about this project. No need for you to find that thread. We can use this one for further discussion about this project.

[Priyank Shankar]

Ohh Thanks, I thought maybe we had more info in that thread

[Akshit Sinha]

Greetings James. Can you elaborate more on how exactly is leveraging any of OpenAI's possible or even practical. As far as I am aware most of OpenAI APIs are proprietary and paid(or only free upto a certain extent). OpenAI's capabilities are great and everything, but I believe the functionality of something like a converter shouldn't be dependent on any external entity and should happen totally offline. Am I missing something? I would love to know your thoughts on this. Thanks! :)

[James Ward]

Yeah, those are good points and maybe it would have to be done manually without external dependencies.

[Clocks]

You don't need a machine learning model for a decision tree

[nicognaw]

I agree that LLMs are powerful and fascinating, and I have personally developed several applications using the OpenAI API. However, I don't believe that we can rely solely on LLMs for all tasks, particularly for complex tasks such as building compilers. Explicit programming may be better suited for these types of tasks. While it may be acceptable to create a demo to demonstrate how LLMs can be used, it is unlikely that it will be accepted as a plugin for the proposed project.