I have yet to have such a moment. To me it is still just a compressed database.
Though I am surprised at how these databases turn professionals into amateurs, like when Meta publishes some chatbot that can trivially be queried into sending account resets to any email address or when large corporations just dump their entire secret sauce into some remote SaaS led by obviously kooky people.
It's like established pros and big corps want to experience what it was like to be a self-taught PHP coder in 2007, like some kind of false nostalgia.
> like when Meta publishes some chatbot that can trivially be queried into sending account resets to any email address
Or when the Director of Alignment at Meta’s Superintelligence Labs ran OpenClaw and it deleted her inbox.
For me that was an "oh shit" moment in the sense of yet again being disappointed by humans not taking AI risks seriously.
Like, come on, how do these people somehow not know that software has bugs? And that AI is harder to debug than basically everything else?
And yet, somehow, there's still a lot of people who think AI can't possibly cause severe problems despite people like this doing things like this with AI like this.