OpenAI shifted safety from messaging to deployment

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 7:43

   There must be one product or strategy you guys have – what's the one that you've disagreed about the most vehemently? I was just thinking when Greg was talking about this. This is not a product, but it was the thing that came to mind that I was going to say even before you asked that. uh we used to talk a lot about how to talk about safety we never disagreed on the extreme importance of safety and what it will mean to get this right or get this wrong but the the field has had a strange relationship with how we've talked about safety how we've used safety and how much that becomes about power versus actually keeping things safe and i i think earlier in our history I got swept up more in the, we got to really talk about this in a particular frame. And Greg was very disciplined about, we're not going to fall into the traditional frame. We can't talk about

2. 8:37

   that way. Now, even then, I think we have, because this is so important, we, I would ever have fallen into the trap of still talking more in the wrong frame than we should. But I think one of OpenAI's greatest contributions to date has been finding a different way to talk about safety not just in how we build the products and how we talk about society needs to do but like what how we deploy them the idea whole idea of iterative deployment and actually getting to maybe not the actually get into a world where we're figuring out how to deploy products that get increasingly safe as the stakes go up and greg really held a line there that i think has been quite important to the company against extreme pressure not to do that and i think that's been like quite quite important to our whole strategy not just how we talk about

3. 9:34

   things but how we ship and build products yeah and if you look at for example the open ai foundation which is the non-profit that governs open ai and has a very large chunk of equity one of its pillars is ai resilience what that really means is thinking about how do we make ai be something positive for the world and the answer is not any one intervention right it's not you have chain of thought monitoring and now you've achieved the mission it's really a whole deep sequence of different ways that society should orient around this technology and i think to this perspective of you're not going to solve AGI going well for the world in a paper. It has to be a worldwide effort from contributions from society, from many different people, from many different ways of really understanding what this technology is, how it will affect people, how it

4. 10:29

   will affect the world. And this is something that was not, I think, at all appreciated or understood when we were starting out 10 years ago, because it's very easy just to fix it on, you know, building technology, that's the only problem we need to solve. I'm not saying anyone explicitly said it that way, but I think sometimes you can fall into a mental trap of thinking about it that way. And so a lot of what I think we have spent time on is being first principles thinkers and really thinking about how do you operationally deliver transformative technology to the world in a way that is going to actually help people in their daily lives. And one thing you realize is, well, if you have a very powerful piece of technology that will change things, probably it's going to go better if you've had a less powerful piece of technology that you've already

5. 11:12

   helped change things in a positive way. And so if you just think about it that way, you start to really be pulled down this road of thinking about resilience, thinking about iterative deployment. And again, I think this is a lot of the dynamic within OpenAI between the two of us that we're always thinking about these questions of how do we actually achieve this mission and make it go better. Yeah. It feels like just yesterday, I saw you at South by 2022, 2023, and it was a