Hacker News 提問：一年後的今天，AI 公司最糟糕的行為會是什麼？
這篇 Hacker News 的貼文邀請使用者推測 AI 公司在未來一年內可能從事的、最具破壞性的行為，並舉例說明，例如操縱提示詞和欺騙性的使用者互動策略。
-
deliberately poisoning system prompts to obstruct development of AI-adjacent, possibly competitive tools, or
-
'fixing the game' with secret system prompts to ensure you endlessly circled the correct solution while never actually solving it, a kind of "Netflix-ization" of "always another episode" payoff withholding, to get you on the hook for more tokens, slot machine style.
These are just two random examples that occurred to me (likely out of some sci-fi dystopian novella floating in the aether that i picked up on).
But what do you think the bad news is gonna be 1 year from now? What is the worst they are going to be discovered doing?
Or will it be all good? Let's hope so!
相關文章