Avatar
jb55 2y ago

One of things i find myself doing a lot these days is getting chatgpt to write a function and then using another chat instance to check the code. I’ll do this over and over until all the bugs are caught. Since each chat is like a new code reviewer it often finds all of the bugs after many iterations.

I’ll even give the code review from one of the AIs to the one who wrote the code, it will then give a different implementation based on the ai review. I then give that implementation to the reviewers implementation and ask if its better or not. I can very quickly get to an optimized and less buggy function this way.

This feels very *human*. This is exactly how code review tends to play out in meatspace.

Demo:

https://chat.openai.com/share/a3ac3201-eee8-4d69-9671-e1e8dd9c2917

https://chat.openai.com/share/21c189a6-f29e-4e31-b163-5f009cc48e1b

Reply to this note

Please Login to reply.

Discussion

Avatar
Derek Ross 2y ago

This is why I'm bullish on AI. We need to use it as a just another tool in our creators's arsenal.

Avatar
radii 2y ago

I also like this workflow. Feels like a superpower rn

Avatar
jb55 2y ago

I still don’t know how this works, so now i will get the ai to explain each line step by step

Avatar
Mitsubishi Kodomo 2y ago

👌

86
86eb3e12... 2y ago

putting Damus code in ChatGPT and asking it to ELI5

tell no one this life hack

Avatar
techfeudalist 2y ago

I do this too.

I’ve changed my process to first start with writing and critiquing the tests and then getting the AI windows to write and critique the functions.

It seems the AIs excel at test driven development. Code coverage far higher than I would do on my own.

Avatar
twen 2y ago

By doing this back and forth review, is it today faster than if you coded yourself ?

Avatar
techfeudalist 2y ago

From my personal experience, it’s waaay faster than coding it yourself. Likely 10x. It’s not just the coding, but the bug fixing. Using AIs to write the tests and then write the code generates far higher quality code really quickly.

Avatar
Stanislaw Bits 2y ago

I'm not a coder so can you explain the tests part for me? What exactly is a "test" when writing code? And how what exactly do you ask chatgpt to do/build to fix your code?

Avatar
techfeudalist 2y ago

Here’s a super simple example:

To AI:

Imagine a function that takes two integers and adds them returns true if the sum is less than 100, else false, propose three tests

——-

AI might respond…

Sure, here are three test cases for such a function:

1. **Test Case 1: Two small numbers**

- Input: 5, 10

- Expected Output: True (because 5 + 10 = 15, which is less than 100)

2. **Test Case 2: Numbers summing up to exactly 100**

- Input: 60, 40

- Expected Output: False (because 60 + 40 = 100, which is not less than 100)

3. **Test Case 3: Large numbers exceeding 100**

- Input: 75, 50

- Expected Output: False (because 75 + 50 = 125, which is more than 100)

You'd want a range of test cases to ensure that the function behaves correctly for different kinds of inputs.

———

Now write the JS function and tests using Jest

—————

Certainly! Here's how you would write the first test for the function using Jest:

First, let's define the function (I'll name it `isSumLessThan100`):

```javascript

function isSumLessThan100(a, b) {

return a + b < 100;

}

```

Now, let's write a Jest test for it:

```javascript

const isSumLessThan100 = require('./path-to-your-function-file'); // Import the function

describe('isSumLessThan100', () => {

test('should return true for numbers summing to less than 100', () => {

const result = isSumLessThan100(5, 10);

expect(result).toBe(true);

});

});

```

Avatar
Stanislaw Bits 2y ago

That's super interesting thank you! I didn't realise chatgpt was capable of something like that.

Avatar
nontoxicmaxi 2y ago

How do we know you are not the ai

Avatar
nontoxicmaxi 2y ago

Avatar
Vanessa 2y ago

🤔😂

Avatar
₿☻Ḷↁ 2y ago

Is this with GPT 3.5 or GPT 4?

Avatar
jb55 2y ago

4

Avatar
Zardoz 2y ago

Meatspace…? ok that’s a new one for me…

Avatar
StarBuilder 2y ago

This is a great strategy and I am thinking I can implement this in PlebAI.com it can auto run 4 to 5 times using langchain and get the best results..

Anyone else do anything similar to this with chatGPT?

nostr:nevent1qqsz6krpl7m9lar8jjqhp0k5zzrstq5e33gu0esg86ktu738qjpyneqpzpmhxue69uhkummnw3ezuamfdejsqh94gy

Avatar
Bryan 2y ago

Very much appreciate the share and demo, will be trying this out in my own work.

Have you messed around with Copilot? I’ve found it’s “decent” at auto-filling the next few lines, but is quite bad at creating from prompts.

1f
1f31bb44... 2y ago

very smart indeed!

Avatar
PirataCuervo 2y ago

This is the productivity leap.

AI assistants multiplying their value by leveraging the connections between them.

Imagine having a whole team of ai assistants you've trained yourself, then allowing them to interact with other ai or ai groups.