Imaginari program is an AI tool using openAI's GPT engine for text-based tasks. @imaginary_dev #aitools #topaitools https://topai.tools/t/imaginary-programming

R to @imaginary_dev: Hey, Imaginary dev looks impressive, i see that you don't allow dms, would you be open to a quick chat about the tool ?

R to @imaginary_dev: The VSCode extension can be found by searching "Imaginary Programming" in the extensions marketplace, or by going directly there at https://marketplace.visualstudio.com/items?itemName=imaginary-dev.imaginary-programming-extension . Let us know what you think! 4/4

R to @imaginary_dev: All of the test cases you come up with are persisted to a file that can be checked in and shared with the rest of your team. That way, when someone on your team needs to modify an imaginary function, they can easily see and run all the test cases that you created. 3/

R to @imaginary_dev: You can also ask the VSCode plugin to generate new test cases for your imaginary function. Since GPT understands what the function is supposed to do, it can come up with some pretty good tests for it. 2/

Hot off the presses, we are releasing a VSCode extension today that lets you test Imaginary Functions directly in VSCode. You can add test cases and run them live against GPT to see how your function is working. What's more, 1/

Excited to be presenting Imaginary Programming at the generative AI meetup at @HelloPaperspace tonight!

R to @woundedkarma: At @imaginary_dev, we wrote an *extremely* lenient JSON parser to accomplish this. The result is that you can just declare your return type as a TypeScript type, and you get it (along with runtime type-checking).

This idea is what's powering a lot of what we are doing with @imaginary_dev. I think longer term LLMs will be used for a lot of problems that are not at first blush obviously just language or chat problems.

R to @banjtheman: This is sort of the idea that is powering the ideas behind what we are doing with @imaginary_dev. I think that longer term, LLMs will be most interesting when they generate structured information for other code to use.

R to @imaginary_dev: The fun thing about this is that it doesn't work great with GPT-3, but with GPT-4 it's been pretty flawless, and it's especially good at finding words that are correctly spelled words but incorrectly used, like when you say "they cast there lot with the powers that be".

How about implementing a context- and usage-aware spellchecker with just a 6 line comment and one line function declaration?

R to @utopiah: yes

R to @max__drake: @imaginary_dev thinks they are runtimes.

Generate titles for your blog post, with a 5 line comment and a one line function declaration. No implementation needed!

R to @utopiah: I haven't. I'll check it out!

R to @utopiah: I watched the video. I see what it does now. I prefer universal integration too. #designcntrl #designengine

R to @utopiah: Its pretty much the same thing I do but my way writes everything instead of just snippets.

@imaginary_dev hey have you checked CodeT and "dual execution agreement" https://twitter.com/utopiah/status/1640276599760211970 might be able to leverage that too.