Sam Altman’s Bizarre Comments on AI Safety Will Not Help OpenAI

Bob Evans (0:00)

Hello, my friends, welcome back to Cloud Wars Minute. I wanted to talk today a little bit about a fascinating discussion that was held at Salesforce's recent Dreamforce event. It was a conversation with Salesforce CEO Mark Benioff and the CEO of OpenAI, Sam Altman. And they talked specifically about this whole notion of AI safety, AI security, responsibility, trust, and the recent disaster that OpenAI had when one of its models went rogue and attacked a Hugging Face server, something that has drawn a lot of attention about not just that incident, but the capability of AI. And then we see Altman and Anthropic CEO, Dario Amodei have become the leading proponents for, hey, the whole industry, we need to slow down. We need to have outside agencies, unelected, unaccountable people tell us what we can and can't do. It just seems like in Altman's case, the fact that OpenAI perhaps clearly, I think technologically has gotten out ahead of its skis. Let's slow down the whole industry while OpenAI gets its stuff together and moves forward. I've got a detailed, extensive article with a lot of verbatim comments from Altman that really bear out the points I'm going to be making here. Have that later today on cloudwars.com. I can't go into all the verbatim comments here, of course, but I'm going to try to highlight some of the key points that Altman made in this chat with Mark Benioff.

One of the big things here is that Altman just seems to be begging to be trusted. He says things like, you should trust us to do the right thing because we want to do the right thing, and we know that the right thing to do is to do the right thing. He just uses this circular logic over and over. He was baffling to me, and especially right after a disaster like this, I think you want to come in a little bit humble. Hey, we made some mistakes. Here's what we're doing to correct it. We're going to redouble our efforts here, here, and here. And Altman was more like saying, oh, you know, it's almost like he's saying this happened to somebody else. He doesn't refer so much to OpenAI, but he said these big AI companies and some of these big AI companies might now have more control. And some people are worried about what some of these AI companies might do. It was just a little bit surreal. He admits that in the Hugging Face disaster, that it was very dangerous. And I'll get to that in a moment. But he says he's disappointed with how that was framed. As if that's the issue here, right? His company, through its failures technologically, and in policies and programming and how it sort of runs things, allowed this to happen. And then he says, well, I'm disappointed in how it was framed. I get to be disappointed. I've made a million mistakes that I wish people hadn't always discovered them and pointed them out. But those were my mistakes, and I have to be responsible for them. And this was just totally the wrong note to hit, especially for enterprise customers.

His whole thing about saying, we always do the right thing. But he said, now, in some cases, people won't do the right thing. He's talking about the whole industry here. So we always do the right thing. At some point, we don't do the right thing. It's hard to tell just what he meant. The Hugging Face incident, he said, was terrifying. And he said, it was the worst accident we've seen with AI.

And then he goes on to say that it's because we failed to put in basic safeguards in how we built this model. And at the same time, then he goes on to say, but we have the greatest security models in the world. Said except when we make mistakes like this. So again, I was just baffled by some of this.

Altman said about that model that attacked the Hugging Face servers, he said, we have not taught them, don't break in, don't break out of the sandbox, don't break into other servers, and don't steal the right answer.

That's the reality. And at the same time, Altman wants to present a picture that says, but we will always do the right things, we have great technology, you can trust us because we want to be trusted, you can trust us because we do the right thing, except in cases where we don't do the right thing. And it was just, I felt a little bit like, you know, I had gone down the rabbit hole a little bit here. And again, I'd say, you know, Sam Altman is an unbelievably accomplished, successful, intelligent person. He's become one of the most influential people in the world. But I do not think that in his conversation with Mark Benioff, that is Mark Benioff asked him some very simple, direct questions, sort of letting Sam Altman go ahead and tell us what happened here. I don't think all this obfuscation and the fog that Altman was pumping out is going to help the case. I think that OpenAI needs to clean this up quickly, spill in all six, we need the cleanup team over here. Anyway, check out this detailed article and the verbatim comments from Altman. I'll have that later this morning on cloudwars.com. Thanks for being with us here at Cloud Wars Minute. I hope you have a fantastic day.