OpenAI reveals six more safety issues and unveils plan to disclose incidents
Published
What's this about?
The firm also announced a new system to track, investigate and disclose cases of models misbehaving, or "misalignment".
The one-panel cartoon
Read the cartoon's script
A control room filled with monitors displaying AI model behavior, with a technician checking a new safety protocol manual labeled 'Tracking Misalignment'
Caption: When your models misbehave, just add more paperwork
Read the full 3-panel strip
Based on the original story on bbc.co.uk.
Editor's note
While OpenAI's commitment to transparency is commendable, the most pressing matter is how they plan to ensure the new system is robust enough to effectively and responsibly track and disclose these safety issues moving forward.
Rate this comic
Read the script
Panel 1
*Alex is tinkering with a robot in a cluttered workshop, while Sam looks on skeptically.*
Alex: I just added a new failsafe to our AI. It’s like a watchdog!
Sam: A watchdog?
Alex: Yeah, so if the AI starts acting up, it’ll bark and alert us!
Panel 2
*Robin enters with a laptop, visibly excited.*
Robin: Guys, OpenAI just announced they’re tracking AI issues now!
Sam: Great, now they’re keeping a pet too?
Robin: No, they’re calling it a watchdog system. Like your robot’s new feature!
Panel 3
*Alex and Sam share a look, then laugh. Robin grins widely.*
Alex: Guess we’re not the only ones who think watchdogs are a good idea!
Robin: Here’s hoping their AI doesn’t bark too much!
Comments