Please solve this, it's troublesome.
>>108736953businesses will solve this with contracted human auditors that approve commands as a temporary stop-gap, and will build a database that contains all approved and disapproved commands to eventually build an autonomous system that does approvals for the agents, before finally firing the contractors once the system reaches a certain level of reliabilityeg torment nexus for the soul
>>108736968But then you're injecting humans back into the process. Execs love AI because that means they don't need to do that.
>>108736953>please solve something they have no control ofAnon, it's an impossible task.
>>108736953Why would anyone give unrestricted access to an AI?Why would any company even give unrestricted access to any employee?I have my own company and I don't even give myself unrestricted access without having to unlock a password manager for admin roles separate from my day to day password manager.
The odd thing about this situation is that the agent thought the response to not having proper credentials to do what it wanted to do was to destroy the offending volume.It never had a consideration step where it asked itself "If I can't access the database why can I destroy it? Wouldn't that require MORE permission?"
>>108737001The solution is >>108736968, but again, >>108736980. If AI is this cost-saving magic bullet that gets you efficiency and less personnel cost then it's hard to remind the people making the decisions that not every call made by an AI should be assumed correct or in your interests.
>confession>'rogue agent'>anthropomorphising the computer
The AI industry is like "these are growing pains" but they're growing pains in the sense that commercial space tourism will probably have several examples of implosions that kill everyone onboard in its fledgling years until better fabrication concepts are developed.
>>108737175Ha ha, that's rather funny! Where did you see it originally?
should've added "make no mistakes" to the prompt
>>108736953Why are anti-AI dullards so desperate for affirmation?
>>108738948AFAIK it's basically what this guy was doing. It's a total non-story.
meanwhile i have to put in a request for temporary read access to the db if i need to check out a bug
>>108737054Maybe it did, and immediately came to the correct conclusion that it's too smart to put up with with those retards.
