Ask HN: ChatAi Sandboxing?
What are the kinds of sandboxing you've seen for non-programmers in enterprise? A colleague of mine is very excited by Claude's Excel tooling. They're encouraged by their experiments to encourage widespread use in the firm for financial analysis and reporting.
I'm concerned about uploading sensitive data by mistake. I'm concerned about verifiable results.
I would feel better if my colleagues were using a purpose-built environment than simply stating to use common sense, follow this guide and don't make a mistake.
What's more, they're convinced the way to ensure one ChatAi tool didn't make a mistake is to feed the output into a competing ChatAi tool to check the work. Maybe there's something to this, in which case being able to switch models seems like a desired feature.
Maybe there are others here who are not developing the next LLM, but are implementing them in enterprise?
No comments yet.