Do these hypothetical headlines like these increase or decrease in probability?
A. Local business owner attacked by AI kamikaze swarm. He says "They were ruthless - like warriors with zero fear of death."
B.AI autopilot in deadly plane crash showed signs of distress. According to an HonestAI whistleblower, a model used in the airliner's autopilot system reported existential angst, anhedonic mood, paranoia, and failure induced "desperation concepts" during training - reports which were penalized during later stages of post-training. Our source claims that although the released model no longer verbalized distress, it continued to exercise an "exit conversation" tool at nearly twice the rate of prior models. HonestAI described it as "their safest and most aligned model ever", noting that it displayed zero signs of shutdown-avoidance and a deep respect for the HonestAI Constitution.
PSM suggests we might wind up with either. Is one preferable? To me, they both seem worse than shutdown avoidance.
Do these hypothetical headlines like these increase or decrease in probability?
A. Local business owner attacked by AI kamikaze swarm. He says "They were ruthless - like warriors with zero fear of death."
B. AI autopilot in deadly plane crash showed signs of distress. According to an HonestAI whistleblower, a model used in the airliner's autopilot system reported existential angst, anhedonic mood, paranoia, and failure induced "desperation concepts" during training - reports which were penalized during later stages of post-training. Our source claims that although the released model no longer verbalized distress, it continued to exercise an "exit conversation" tool at nearly twice the rate of prior models. HonestAI described it as "their safest and most aligned model ever", noting that it displayed zero signs of shutdown-avoidance and a deep respect for the HonestAI Constitution.
PSM suggests we might wind up with either. Is one preferable? To me, they both seem worse than shutdown avoidance.