Skip to content

After using Copilot ... meh, it's still a lot like the helpful little MSFT Clippy or Intellisense autocompletion, but it will get better ... when it becomes ubiquitous or something like indoor plumbing, sewer, electricity and HVAC. #33

@MarkBruns

Description

@MarkBruns

We are indeed using Copilot ... actually the more edgey Copilot Nightly with Copilot Labs ... one really can see value from here ...

It will eventually get there ... particularly for the developers who really work on their own skills and really practice using it ... really using a good debugger and committing extra effort to writing cleaner, tighter, more maintainable code might take developers further ... but AI is trendy ... probably, the familiarity with it helps in interviews and gets people jobs.

In the case of AI-assisted coding, really using means 10,000 to 20,000 hours of intent, informed use in which one thinks in AI-land ... muscle memory takes over and the developer doesn't even really think any more at all about using something like Copilot. Really using something is all about getting to the point where it just works and is completely taken for granted, ie like how people use the shitter and just expect the light in the bathroom to work, the water to have filled the toilet tank and the sewer drain to be free of tree roots

We know that it is still very early in the game ... Copilot or OpenAI's CODEX are still a lot like Windows 1.0 or maybe pre-Windows 0.X from 40 years ago ... mostly a curiosity for early adopters who excited about how a graphical user interface might make using a computer different than relying upon fluency in the MS-DOS prompt.

Copilot [and GitHub] are still extremely Microsoftey ... especially in terms of GitHub's Ballmeresque exuberance ... Steve and Bill are the Rolling Stones and CoPilot is probably not going to be a legit peer-programmer, but that does not negate the importance of Win95, MSOffice, VisualBasic and all of the vestiges of MSFT that we still live with, even if we do not listen to the Stones every day.

The POINT is that the ubiquity of things that people can COUNT on JUST WORKING really does matter.

So we need to get past the headlines in the news and find out a LOT more about really USING Copilot powered by the OpenAI Codex model, AWS CodeWhisperer ... especially Google Research's PaLM / Pathways

https://ai.googleblog.com/search/label/Natural%20Language%20Processing?max-results=11

https://ai.googleblog.com/search/label/Semantic%20Models?max-results=11

https://ai.googleblog.com/2022/08/announcing-patent-phrase-similarity.html

https://ai.googleblog.com/2022/07/ml-enhanced-code-completion-improves.html

https://www.connectedpapers.com/main/45263786d07f5751f7494fdeee3c8764836d02c4/NatGen:-generative-pre%20training-by-%E2%80%9Cnaturalizing%E2%80%9D-source-code/graph

https://www.connectedpapers.com/main/7b46b9da287429d19a00ca3f9219c1020f7c9df8/A-Survey-on-Artificial-Intelligence-for-Source-Code:-A-Dialogue-Systems-Perspective/graph

https://www.connectedpapers.com/main/acbdbf49f9bc3f151b93d9ca9a06009f4f6eb269/Evaluating-Large-Language-Models-Trained-on-Code/graph

https://www.connectedpapers.com/main/512e9aa873a1cd7eb61ce1f25ca7df6acb7e2352/Pathways%3A-Asynchronous-Distributed-Dataflow-for-ML/graph

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions