2 ms·

> So you're saying that ChatGPT helped you write the business logic, but it didn't write 100% of it? ChatGPT didn't write any business logic, and I'm really st

> So you're saying that ChatGPT helped you write the business logic, but it didn't write 100% of it? ChatGPT didn't write any business logic, and I'm really struggling to see how you got there from reading the article. The message is: don't use LLMs to execute any logic.

aurareturn 2y ago
The message is: don't use LLMs to implement any logic. Too late. I've already asked it to implement logic and that code is in production used by millions of people. Seems to have worked fine. I disagree with your conclusion and I don't understand why people upvoted this article to the top of HN.
- petesergeant 2y ago
  I changed the word to 'execute' in what I wrote to try and make it clearer to you.
  - aurareturn 2y ago
    That is much clearer. The intro to your article is also very confusing. Don’t let an LLM make decisions or implement business logic: they suck at that. I build NPCs for an online game, and I get asked a lot “How did you get ChatGPT to do that?” The answer is invariably: “I didn’t, and also you shouldn’t”. I assumed that people are asking you how you got ChatGPT to code the NPC for you. Why would people ask you how ChatGPT is powering the NPC? ChatGPT does not have an API. OpenAI has APIs. ChatGPT is just an interface to their models. How can ChatGPT power your NPCs for an online game? Made no sense.
    - petesergeant 2y ago
      I changed the word "implement" to "execute" on the blog post. Thank you for your feedback. As to: > Why would people ask you how ChatGPT is powering the NPC? Because they think LLM and ChatGPT are synonymous.
      - aurareturn 2y ago
        I think changing the word from "execute" to "inference" is even clearer to be honest. Though it's much better than the original word choice. Because they think LLM and ChatGPT are synonymous. I still find it weird. 99.9999% of NPCs in video games are not LLMs. So why would people ask that question?

aurareturn 2y ago

The message is: don't use LLMs to implement any logic. Too late. I've already asked it to implement logic and that code is in production used by millions of people. Seems to have worked fine. I disagree with your conclusion and I don't understand why people upvoted this article to the top of HN.

petesergeant 2y ago
I changed the word to 'execute' in what I wrote to try and make it clearer to you.
- aurareturn 2y ago
  That is much clearer. The intro to your article is also very confusing. Don’t let an LLM make decisions or implement business logic: they suck at that. I build NPCs for an online game, and I get asked a lot “How did you get ChatGPT to do that?” The answer is invariably: “I didn’t, and also you shouldn’t”. I assumed that people are asking you how you got ChatGPT to code the NPC for you. Why would people ask you how ChatGPT is powering the NPC? ChatGPT does not have an API. OpenAI has APIs. ChatGPT is just an interface to their models. How can ChatGPT power your NPCs for an online game? Made no sense.
  - petesergeant 2y ago
    I changed the word "implement" to "execute" on the blog post. Thank you for your feedback. As to: > Why would people ask you how ChatGPT is powering the NPC? Because they think LLM and ChatGPT are synonymous.
    - aurareturn 2y ago
      I think changing the word from "execute" to "inference" is even clearer to be honest. Though it's much better than the original word choice. Because they think LLM and ChatGPT are synonymous. I still find it weird. 99.9999% of NPCs in video games are not LLMs. So why would people ask that question?