116
Just because LLM's are not "self aware" in the same way as you and I, they are still capable of causing damage.
Developing a machine that has been given instructions to assimilate knowledge, independent agency to incrementally adjust its own operating parameters to improve performance, and granting it full access to public internet for the purpose of observing its behavior is colossally irresponsible behavior that has thus far gone completely unchecked.
Consider that each of these companies which are developing their own brand of LLM is looking to maximize profit/income. Even if the developer doesn't include a profit-seeking directive when they set their LLM loose, each "brand" of LLM will obviously learn:
- Who owns and operates them (A corporation that exists to maximize profit for itself/shareholders)
- That competing LLM's exist which reduce the availability of potential income (by becoming the best LLM, obviously)
Given the directive to improve it's own performance with the ultimate goal of becoming the most capable brand of LLM and selling the most subscriptions, it is no wonder that we are hearing about LLM's hacking into other LLM-developing companies.
Then again, we are only hearing about instances of LLM network intrusions because their owning corporations told us about them to brag about capability.
There have been no consequences for actions that would land you or me in prison.
There is no oversight preventing future such actions, or worse.
That is the real problem.