Thank you Vladimir Putin for saying what needed to be said.
(Open/Gen-) AI is dangerous. Very dangerous! And something needs to be done about it!
“Humanity has to consider what is going to happen due to the newest developments in genetics or in AI. One can make an approximate prediction of what will happen. Once mankind felt an existential threat coming from nuclear weapons, all nuclear nations began to come to terms with one another since they realized that negligent use of nuclear weaponry could drive humanity to extinction.
It is impossible to stop research in genetics or AI today, just as it was impossible to stop the use of gunpowder back in the day. But as soon as we realize that the threat comes from unbridled and uncontrolled development of AI, or genetics, or any other fields, the time will come to reach an international agreement on how to regulate these things.”
I don’t know about you, but with respect to what has been advertised, these are the six variants of Open/Gen-AI the doctor sees:
Gender/Race-Biased: especially in HR; it’s trained on “good resumes”, but, guess what, when those “good resumes” were selected from a pool of hired candidates that have predominantly been white men, guess what the AI looks for?
Hallucinatory: too many stories to track now of AI creating fake summaries on fake articles by fake authors for which it created fake profiles; Lawyers have fall for this multiple times!
Harmful/Hateful: train it on open data which contains hate speech, just like a kid exposed to its first profanity, it mimics … non-stop
Murderous: multiple examples of self-help chat systems literally telling people to kill themselves (and then a few examples of people actually doing this) as well as self-driving systems ignoring the “shadows” of what were people RIGHT in front of them
Sleeper: the newest threat, sleeper behaviour that can go undetected for days, months, or years until a specific date or phrase is entered (in combination); the perfect sleeper agent!
Thieving: not only are these open AI plays generally trained on stolen data, but since all your queries and outputs are directly used (or indirectly influence) the network, they steal your data (even when the designers didn’t set about to do so)