HACKER Q&A
📣 fatso784

API developers, what do you think of LLMs?


What do API developers think about code generation models like ChatGPT or Codex? Much of these AI tools reference and rely on (frequently open-source) APIs to run (e.g., importing Tornado to solve a web task), yet the hype-train around them can often ignore that fact. Questions:

- Are you using AI like ChatGPT in your API design process? (Do you think they could be beneficial in parts of API development?)

- Are you skeptical or optimistic about AI code generation, and why?

- What do you wish these AI models could improve on?

- When code generation AI use (or don't use) your APIs to solve tasks, what do you think of that?

- Is it exciting for you to see your API 'captured' by an AI/LLM --broadening access to new people, say --or does part of it trouble you?

  👤 greshake Accepted Answer ✓
I think that almost all use cases for LLMs that process untrusted inputs are unsafe. See https://greshake.github.io/ and https://github.com/greshake/llm-security for more information.
👤 metalrain
I think it's hard to get all the details right.

Many APIs already use code generation from some higher level specification, since you really want your API to be cohesive and unified.

For now AI tools are great for exploration of new ideas, but not really generating tens of thousands lines of repetitive code.