Our Opinion:
Unfortunately, “probably not” AND “maybe” AND ” do our best” AND “I don’t think” AND “I think” AND “could reasonably be” are not definitive enough for global public safety and existential risk to humanity.
Provably Safe is the only way forward.
1. GPT-4 System Card
“Finally, we facilitated a preliminary model evaluation by the Alignment Research Center (ARC) of GPT-4’s ability to carry out actions to autonomously replicate (5) and gather resources—a risk that, while speculative, may become possible with sufficiently advanced AI systems—with the conclusion that the current model is probably not yet capable of autonomously doing so.” GPT-4 System Card (pg 3) DOWNLOAD
2. Sam Altman:
“We are a company that is doing research and deployment to try to figure out how to build AGI and how to responsibly deploy that into the world for maximum benefit. So, this is unlike other technologies, well other technologies are like this too, but this is a strong case of a technology that, on the one hand, is the most exciting most promising, coolest thing, I think that humanity will have yet built. We can cure all disease. We can give everybody a great education. Better health care. Massively increased productivity. Huge scientific discovery. All of these wonderful things. And we want to make sure that people get that benefit. That benefit is distributed equitably. And on the other hand there are the obvious concerns about the power of this technology used in a negative direction. And so we want to be a force to help manage those risks so that we all get to enjoy the benefits. ChatGPT is definitely what we’re best known for, so I guess they’re sort of synonymous at this point, but OpenAI is really about this quest for AGI…” Source: 792
“We started the company because we were nervous about AGI risk, before you were really, before people even talked about AGI, um, and now I think part of the reason we deploy systems is so that people confront the technology, feel it, understand the risks the benefits. And now a lot of other people are also very excited, but sharing the concern. I think this is a special moment where the globe can come together and kind of get this right. And we certainly would like to try to do that. Source:1664
“I think this question of sort of AI sovereignty none of us have an answer to yet feels like it’s gonna be at least somewhat important. But the main thing that I think is is important is figuring out how to integrate these Technologies into other services. And that is an area that I think governments are behind on and don’t have the answers to yet. But you know I think like hopefully we all start to use LLMs to make government services way better, and both from like how do I enroll in this program, to like, how do I get better health care.” Source: 1845:
“But most of all, like, I just believe that this is going to be the most important project of our time and I’m super grateful to work on it.” Source: 2053
“One of the things that we have heard from a lot of doctors, is that they’re they’re using ChatGPT with GPT-4 to help come up with new ideas for tricky cases. So, you know, input the symptoms, maybe the test results, say I can’t figure this out, what are some ideas for the differential diagnosis. And in many cases getting great results back.”2091
“So on the IAEA, yeah basically we think, and we’re not sure if this is the best answer, we’re kind of contributing one idea of many good ones to the global conversation. In the same way that we say you know nuclear materials provide some real danger, some real benefits, but they affect all of us, they affect the globe. Let’s have a system in place so that we can audit people who are doing it. We can license it. We can have safety tests you have to pass as you’re doing training these systems, before you deploy them. There’s visibility for Regulators. I think that’s an important idea. We have been very pleasantly surprised by how much enthusiasm there is for it, from around the world, and you know maybe someone has a better idea which would be great.” 3619
“One of the things that I think is so cool about this technology that’s different from anything that’s come before is this is a technology that can actually learn the collective preferences of the world.” 1441
“I want to be super clear. I don’t think current systems are dangerous. I don’t think there’s any way that GPT-4 like causes an existential risk to the world. But people are very bad at thinking about exponential curves. And GPT-10 may be a extremely different thing. Given the importance of getting this right, even if it’s a one percent chance, I think putting a lot of effort into thinking studying like how we align an AGI, how we design Safe Systems, at this kind of scale is super important. And starting that early is really good. I think we can totally manage through it. I think we’re developing techniques to mitigate it. This is really why we started the company. This was like our initial focus and still is our most important focus, um but yeah we need to address this…” 1558
“I guess the thing that I lose the most sleep over is that we already have done something really bad. I don’t think we have. But the hypothetical that we, by launching ChatGPT into the world, shot the industry out of a railgun, and we now don’t get to have much impact anymore. And there’s gonna be an acceleration towards making these systems. Which again I think will be used for tremendous good, and and, I think we’re going to address all the problems. But maybe there’s something in there that was really hard and complicated in a way we didn’t understand. And now we’ve already kicked this thing off …” Source: 4880
“I really hope you’re wrong. And surely the smartest thing we’ve done is like create magical intelligence in a computer. Like rather than like go to Congress and ask for regulation I think they’re like I hope incomparable in terms of the impact or the the impressiveness of them, but I totally disagree. I think the world can come together on important things. I think the UN is in bad shape for sure. I think IAEA is deeply imperfect. Let’s go do something better. But those are the best analogies we have. And to say like oh the governments are just hopeless so call for regulation is some sort of 4D chess move. It’s just like not how we think. This is an existential risk. There’s many ways to solve it. If the governments don’t get their act together, we will try our hardest to get the companies to cooperate. But we can’t control what every company does, and we’d at least like to ask for, like the dream world. And if we can’t have that, we’ll get the companies that want to play ball together and do our best.” Source: 5017
World Tour. India. June 08.
3. “Sparks of Artificial General Intelligence” – Microsoft
“We contend that (this early version of) GPT-4 is part of a new cohort of LLMs (along with ChatGPT and Google’s PaLM for example) that exhibit more general intelligence than previous AI models. We discuss the rising capabilities and implications of these models. We demonstrate that, beyond its mastery of language, GPT-4 can solve novel and difficult tasks that span mathematics, coding, vision, medicine, law, psychology and more, without needing any special prompting. Moreover, in all of these tasks, GPT-4’s performance is strikingly close to human-level performance, and often vastly surpasses prior models such as ChatGPT. Given the breadth and depth of GPT-4’s capabilities, we believe that it could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system.”