Comment by dahart
2 years ago
Does the crystal ball say anything about what’s going to happen if Copilot or Bing start revealing non-public information to anyone who asks? It’s bound to happen if they train on non-public information. Imagine Microsoft accidentally releasing other companies’ corporate strategies and proprietary internal tech, or people’s personal finances and private social interactions. I would foresee both major litigation and government regulation coming down pretty hard. I would also expect a dramatic migration away from the product if something like that came to light. I honestly hope they’re smarter that this- training on data without explicit permission is already one of the biggest problems with AI efforts.
You use public data only for your external-facing products like Copilot and Bing.
You use all data, public and private, for your in-house skunkworks LLM-AI used by vetted, NDA-bound staff and execs.
Bing won't be able to answer questions like "What are the monthly active user counts for CoolService LLC?" or "What are the manufacturing processes used at Gadgetmaster International?" but maybe DarkBing will.
Even if LLMs aren't good enough to deliver those answers today, they might be in five or ten years, and in the meantime you want to fill the pool of data you're going to feed it.
Cynical speculation? Yes. Eventually possible? Maybe...