My team, Safety Systems, is working on the practical side of alignment at OpenAI: Building systems to enable safe deployment of powerful AI models. Our work encompasses a wide range of research and engineering projects to reduce unwanted use cases and ensure model behavior within our safety standard and legal compliance. If you are ... - keen on defining evaluation standards for deciding whether a model is safe enough to be deployed. - worried about harmful use cases and want to detect and stop them. - interested in training the model to say no to harmful requests and to be robust to jailbreak style vulnerabilities. Then come join us: - https://lnkd.in/g7xSR-Rs - https://lnkd.in/gMsDfUXu
Monica Anderson
6mo
I elucidate my favorite aspect of AI Alignment at https://zerothprinciples.substack.com/p/ai-alignment-is-trivial I'd be happy to discuss that with anyone.
Given safety is a public concern , openAI can setup an open source AI safety institute so all tech companies or adopting AI companies can contribute and share these ideas 💡.
Jonathan Lachman
7mo
Lilian Weng is a rare talent and is leading vitally important AI safety work for us. Come join her!
Qingsong Yao
7mo
What about your team share more about how alignment was done in GPT-4 so that all others can learn from it?
I do not know your email Lilian Weng, but I would love to connect. I appreciate you sharing your blog! Thank you.
Nupur Kantamneni
7mo
Hey, just messaged you!
Founder @ Strategic Advisory for Experience | Senior Design Strategist
7moFor the casual reader lost between benefit and existential threat narratives, it will help if you can clarify these terms - "safe model", "harmful use case", "harmful request." From your vantage, these definitions and accompanying illustrations will be really welcome. If it's published elsewhere, please share a link. Personally, I am curious about the composition of teams at OpenAI who tackle these issues.