No single person or institution should define ideal AI behavior for everyone.
Today, we’re sharing early results from collective alignment, a research effort where we asked the public about how models should behave by default.
Blog here:
- OpenAI is nothing without its people
- i love the openai team so much
- Back to our regularly scheduled coding ❤️We have reached an agreement in principle for Sam Altman to return to OpenAI as CEO with a new initial board of Bret Taylor (Chair), Larry Summers, and Adam D'Angelo. We are collaborating to figure out the details. Thank you so much for your patience through this.
- Thrilled to share this update! Working with these 10 teams to me reaffirmed the promise of better tools and processes to better govern our present and our future. Read about it all in our blog & their reports. Much work remains! I'm hiring researchers to join us on this missionWe provided grants to 10 teams who developed innovative prototypes for using democratic input to help define AI system behavior. Read about our learnings and how we'll put this work into practice. openai.com/blog/democrati…
- Cameroon has 240+ indigenous languages. Imagine my surprise today when I learned that the one I speak with my mom's side is not one of them, but rather a made-up one she and her siblings concocted as kids. Explains so many past blank stares.
- One of the most delightful features of DALL-E 2 is the element of surprise. I most often have no idea how it'll interpret strange queries or concepts I'm trying to convey. A few prompts that surprised and delighted me: labs.openai.com/s/fV3RhC0desjV…
- Replying to @ThankYourNiceAI
- Come pay the alignment tax (or maybe learn about InstructGPT) at Poster #920.
- Replying to @ThankYourNiceAIWe surveyed 1,000 people worldwide and compared their collective view to our Model Spec. We found they largely agree with the Spec, and adopted changes from the disagreements. This was hard because human values are complex, and perspectives on ideal AI behavior differ widely.
- Replying to @ThankYourNiceAIThe changes we adopted through this process will soon be reflected in the Model Spec. We’re also sharing our dataset publicly with the research community:
- Replying to @ThankYourNiceAIHow it worked: Participants ranked responses to subjective prompts and explained their rankings. We compared their rankings and justifications against the current OpenAI Model Spec.
- Replying to @ThankYourNiceAIHuge congratulations to the team @mitchellgordon, @ezhang7423, @SandhiniAgarwal, and all of the incredible folks at OpenAI who supported this work!!
