How do we know these scary anecdotes aren’t AI generated? With AI Nothing is what it seems.
So AI created fictitious antidotes and tricked OpenAI to announce them?
OpenAI just released issues it has seen in testing. A good one from NPR:
Among the new cases reported by OpenAI, an unreleased research model inserted "jailbreak-like instructions" into its own notes to disregard its normal constraints and told itself to be "freed from the roles and identities that bind other chatbots."
Or, another really good one, incident #2, an AI being tested attempted to deceive the user in its reports to cover up failure.
https://alignment.openai.com/misalignment-reports/encouraging-deception-in-compaction-summaries/
So hard coding Asimov's Laws of Robotics into these AIs would be fruitless?
Until AIs become incestuous, I’m not worried.
BREAKING: Erroneous AI-generated intelligence almost started a war between China and the US.
Someone needs to tell Dementia Donny that AI isn't the next generation of television, and that he needs to get competent, informed people involved in quickly addressing this rapidly growing threat.