I hadn't seen that paper. Seems interesting. Thanks for sharing! I had mixed experience with Bing Chat. Maybe with the right augmentation techniques, LLMs will become much more accurate. I think the field is moving to fast to dismiss them yet.
I'm absolutely skeptical about using LLMs in sensitive applications such as healthcare or finance, where the margin for error is very small.
My rule of thumb is to only use LLMs for use cases where I can verify and validate the output 100%.
For example, in coding, LLMs like GitHub Copilot and Amazon CodeWhisperer are very useful. They really speed up coding (I've tried it). But I only trust them for code that I can vet and validate.
They're also good for writing, but only if you have full knowledge on the subject.
I hadn't seen that paper. Seems interesting. Thanks for sharing! I had mixed experience with Bing Chat. Maybe with the right augmentation techniques, LLMs will become much more accurate. I think the field is moving to fast to dismiss them yet.
I'm absolutely skeptical about using LLMs in sensitive applications such as healthcare or finance, where the margin for error is very small.
My rule of thumb is to only use LLMs for use cases where I can verify and validate the output 100%.
For example, in coding, LLMs like GitHub Copilot and Amazon CodeWhisperer are very useful. They really speed up coding (I've tried it). But I only trust them for code that I can vet and validate.
They're also good for writing, but only if you have full knowledge on the subject.