4 ms·

Everyone is talking about chatgpt and code generation this week. I think the code generated with chatgpt might be 90% accurate but how confident would you be to

Everyone is talking about chatgpt and code generation this week. I think the code generated with chatgpt might be 90% accurate but how confident would you be to deploy that in production. We know empirically that last 10% can take decades to get it right. Using gpt to generate a blob of content is different from using gpt to generate code.

tablespoon 4y ago
>> I think low code and AI have an incredible productive future. ChatGPT can create simple python programs right now. Imagine a few years from now, where it can output a low code program; that will be game changing since there won't be a need to have a software engineer to deal with the code. It's coming and it's coming fast. > Everyone is talking about chatgpt and code generation this week. I think the code generated with chatgpt might be 90% accurate but how confident would you be to deploy that in production. I haven't been paying close attention, but all the code I've seen it generate is also very simple -- like the kind of stuff someone could write after taking an intro programming class, prompted by a pretty detailed specification. Modern-day software engineers basically write detailed specifications, they don't translate specifications into "code." Whether you're writing "AI prompts" or Python, it'll still be software engineering. Though my gut feel is software engineering by prompts will feel like typing with a wet noodle to hit the keys.
- debarshri 4y ago
  Most complex one I saw this weekend was this [1], the point I'm trying to make is the even if you make generate complex stuff, if would take decades where we deploy this kind of workloads in real world scenarios. [1] https://towardsdatascience.com/i-used-chatgpt-to-create-an-entire-ai-application-on-aws-5b90e34c3d50 https://towardsdatascience.com/i-used-chatgpt-to-create-an-e...
adolph 4y ago
Maybe it would be better to use AI on the test side of TDD. Have the AI listen to the requirements and write tests. Have the human write code to the tests and update the nonsensical parts of AI generated tests.

tablespoon 4y ago

>> I think low code and AI have an incredible productive future. ChatGPT can create simple python programs right now. Imagine a few years from now, where it can output a low code program; that will be game changing since there won't be a need to have a software engineer to deal with the code. It's coming and it's coming fast. > Everyone is talking about chatgpt and code generation this week. I think the code generated with chatgpt might be 90% accurate but how confident would you be to deploy that in production. I haven't been paying close attention, but all the code I've seen it generate is also very simple -- like the kind of stuff someone could write after taking an intro programming class, prompted by a pretty detailed specification. Modern-day software engineers basically write detailed specifications, they don't translate specifications into "code." Whether you're writing "AI prompts" or Python, it'll still be software engineering. Though my gut feel is software engineering by prompts will feel like typing with a wet noodle to hit the keys.

debarshri 4y ago
Most complex one I saw this weekend was this [1], the point I'm trying to make is the even if you make generate complex stuff, if would take decades where we deploy this kind of workloads in real world scenarios. [1] https://towardsdatascience.com/i-used-chatgpt-to-create-an-entire-ai-application-on-aws-5b90e34c3d50 https://towardsdatascience.com/i-used-chatgpt-to-create-an-e...

adolph 4y ago

Maybe it would be better to use AI on the test side of TDD. Have the AI listen to the requirements and write tests. Have the human write code to the tests and update the nonsensical parts of AI generated tests.