Open AI has people read interactions and fix the bad examples. Published ones are probably first to go. So it is a moving target. Wack-a-mole.
SAFETY
-
Jailbreaking Models or Prompting Humans Effectively?
By
–
Are those models jailbreaking or humans prompting them out?
-
Infinite Space of AI Misalignment Ideas Remains Open
By
–
Agreed but that still leaves open an infinite space of ideas to “misalign” upon.
-
General Intelligence Routes and Ethical AI Reasoning
By
–
More that a truly general intelligence can find its way to any idea through nearly infinite routes. And there are no absolute ethical truths but some are closer to the truth than others and can be reasoned as such.
-
OpenAI Should Incentivize Jailbreaks with Robux for Red Teaming
By
–
if @openai really wanted to red team their models they would incentivize successful jailbreaks with robux rewards and watch as a million 13-year-olds go at it if they don't do this they're not serious abt alignment
-
AGI Alignment: Impossible Control or Trivial Problem?
By
–
If it’s an AGI, it can’t be “aligned” any more than a human can, and if it’s not an AGI, it can be trivially controlled and doesn’t need “alignment.”
-
ChatGPT Internet Access Raises AI Safety Concerns
By
–
A colleague said to me yesterday they didn’t realise ChatGPT wasn’t connected to the internet. That’s now changing. Scientists who think a lot about the risks of AI systems tell me they’re *pretty* concerned about this kind of unfettered info access.. https://t.co/pQpi7m1L5i
— Madhumita Murgia (@madhumita29) 23 mars 2023A colleague said to me yesterday they didn’t realise ChatGPT wasn’t connected to the internet. That’s now changing. Scientists who think a lot about the risks of AI systems tell me they’re *pretty* concerned about this kind of unfettered info access..
-
Deepfakes of Trump’s arrest spread on Twitter via AI
By
–
#deepfakes claiming to show Trump's arrest spread across Twitter #AI #RuleoftheRobots https://
nypost.com/2023/03/22/chi
lling-deepfakes-claiming-to-show-trumps-arrest-spread-across-twitter/?utm_source=twitter_sitebuttons&utm_medium=sitebuttons&utm_campaign=sitebuttons
… via @nypost -
ChatGPT Plugins: Third-Party Integration and Real-World Impact
By
–
We are adding support for plugins to ChatGPT — extensions which integrate it with third-party services or allow it to access up-to-date information. We’re starting small to study real-world use, impact, and safety and alignment challenges: https://t.co/A9epaBBBzx pic.twitter.com/KS5jcFoNhf
— OpenAI (@OpenAI) 23 mars 2023We are adding support for plugins to ChatGPT — extensions which integrate it with third-party services or allow it to access up-to-date information. We’re starting small to study real-world use, impact, and safety and alignment challenges: https://
openai.com/blog/chatgpt-p
lugins
…