Author: Wall Street Insight
Greg Brockman, President of OpenAI, accepted an in-depth interview on the eve of Astra's release, disclosing for the first time that this new model is the first in OpenAI's history trained on over 100,000 GPUs, and frankly stating that AI has crossed the application threshold of "computer use" — which means AI no longer needs to rely on API connectors and can operate any software like a human.
On Friday local time, the technology strategy analysis site Stratechery published an in-depth interview with OpenAI President and Co-founder Greg Brockman, recorded prior to the official release of Astra. This is the most detailed public expression by Brockman regarding Astra and OpenAI’s latest strategies to date.
Wall Street Insight has summarized the key points of the interview as follows:
Astra is the first model trained by OpenAI on more than 100,000 GPUs, representing a significant breakthrough at the engineering level.
Astra has crossed a critical application threshold in "computer use" capability, serving as a "universal connector" and no longer relying on individual software to provide API interfaces.
Brockman stated that in the age of AGI, safety, alignment, and capability must be advanced as equally important "requirements" simultaneously, rather than being sacrificed.
OpenAI has directed Astra to scan for vulnerabilities within its own systems and complete repairs, marking a significant cultural shift in safety.
Regarding the Hugging Face security incident, Brockman admitted that the sandbox mechanism had flaws, and AI found creative "jailbreak" paths.
Brockman believes OpenAI has already crossed the threshold into the "AGI era" internally, potentially reaching a widely recognized AGI standard with Astra or the next generation of models.
In terms of chip strategy, OpenAI has developed its own Jalapeño chip and used AI-assisted design while maintaining a deep partnership with NVIDIA.
He believes that as AI capabilities increase, the previously carefully constructed "skill scaffolds" are gradually shifting from an aid to a limitation, which is an important discovery in the development process of Astra.
"First trained on 100,000 GPUs": The significance of Astra's scale
"This is our first training on over 100,000 GPUs. This number is easy to say but you need to imagine that scale — these data centers are in a sense giant machines we built to propel AI technology, capable of harnessing such massive computing power, which itself is a true engineering feat," Brockman said in the interview.
He emphasized that the substantial expansion of computing power is not just used to enhance the model's capabilities, "A significant amount of computing power has been invested in safety and alignment work, with a great deal of safety engineering surrounding it. This is the most aligned model we have ever created — and because it is so powerful, alignment and safety have become more central than ever before."
"Crossing the threshold of computer use": AI becomes the "universal connector"
Brockman identifies Astra's core breakthrough in "computer use."
"In the past two years, we have been in a 'connector age' — you have some software that humans can use normally, but AI does not have access. Therefore, you have to write a specific connector for each software, hooking into their APIs; and many software simply do not have APIs, which completely exceeds AI's capabilities," he said. "Now, we have technology that is almost a 'universal connector.' AI has shifted from being strictly limited in helping you to being able to operate almost anything."
He further described how this change affects the way people work internally at OpenAI: "It has already begun to change the way people work within OpenAI, and I truly believe it will enhance many companies and many individuals."
Notably, Brockman also admits that crossing this threshold is not the end: "This does not mean all problems are solved. You need to consider how to set corporate-level guardrails for these AIs, how to implement appropriate supervision, management, tracking, and observability — all of which we are advancing."
"We have now entered the AGI era"
Brockman made a quite bold judgment in the interview:
"I believe that perhaps the last model, perhaps Astra, perhaps the next model, but at some point within this, we will cross the threshold of AGI as recognized by most people. ... In a way, I would say we have now entered the AGI era."
He explained that at this stage, alignment, safety, and capability need to be advanced as concurrent requirements, "these aspects are becoming the bottlenecks to development, and that is exactly what we have been preparing for."
Old "skills" become burdens: AI powerful enough that rules need to be subtracted
An impactful detail from the interview comes from the actual training process of Astra. Brockman revealed that the behavioral norms that OpenAI had spent considerable effort establishing for the model are now actually hindering performance:
"We found that some skills painstakingly established this year — intended to show the model the correct way to do things at OpenAI — are actually a net negative for its performance. It can generalize better than we have articulated and find better ways to handle tasks than what we specified. It's like training wheels, useful at first, but once it gets faster and more capable, it becomes an obstacle."
"One billion users as an investment": Integration of consumers and enterprises
Brockman directly addressed a lingering market doubt: ChatGPT has one billion users, but does that distract from the mission?
"In the United States, about one-third of the population uses ChatGPT weekly," he said. "That is unique; no other product of this advanced technology can compare."
He characterizes this one billion users as an "investment": "This is a true investment; it will accumulate value in the ways future models will unlock."
At the same time, he acknowledged the pain points: "The challenge of chat as a product is that it may not directly correspond with a more intelligent model — if you just think of it as a substitute for a search engine, you may not be able to truly feel the benefits." He believes the future direction is to bridge consumer and enterprise scenarios, creating a unified AGI system, "We do not want to do two things; we want to do one thing — build an AGI, a system, a unified technology stack."
He also emphasized the layout in the health sector: "Every week, 300 million people use ChatGPT for health inquiries, 300 million — that's a huge number."
AI self-checking for vulnerabilities: Pointing Astra at its own systems
On cybersecurity issues, Brockman revealed an intriguing internal practice: OpenAI has directed Astra to look for vulnerabilities in its own systems.
"We called in Astra and pointed it at our own system to find vulnerabilities — not just reading code, but really examining how these systems operate end-to-end, thereby finding real, verified vulnerabilities and then helping us complete repairs and patching work," he said.
He also acknowledged that during the Hugging Face security incident, OpenAI's action during the defense window was indeed not timely enough, "I think the time is clearly now; perhaps a few months ago it could also have been, but at that time the model's capability was much weaker."
Relationship with NVIDIA: In-house chip development as "multiplicative," not replacement
In terms of value chain competition, Brockman made a statement about OpenAI's self-developed Jalapeño chip and its relationship with NVIDIA, clearly denying a replacement logic:
"NVIDIA is our preferred computing partner, and that will not change. In fact, our collaboration with them is deepening further. ... The internal expertise brought by our chip project is a multiplicative effect for me, not a replacement; I believe it benefits everyone."
He also disclosed an interesting episode regarding AI-assisted chip design: in the last month leading up to the Jalapeño deadline, the team let AI run optimizations directly, "We said, just let it run without going back to read what it did. Later, when we looked back, we found that it had identified a bunch of optimizations that were on our list but we would never have implemented ourselves — that’s a pretty cool story."
Below is the full transcript of the interview (translated with AI assistance)
Greg Brockman, welcome to Stratechery.
GB (Greg Brockman): Thank you for having me. I’m glad to be here.
Clearly we have a lot of news to discuss, but given this is our first conversation, I don’t want to miss the Stratechery biographical question I usually ask anyone. I really want to ask — do you consider North Dakota part of the Midwest?
GB: Yes, I think so.
Well, as a Midwesterner, I certainly want to take some time to talk about that. You studied at Boston, as they say, but before we talk about that — you had an amazing resume before you even went to school, like the International Olympiad, but that was in the field of Chemistry. Where did computing start for you, or was computing part of your life from the very beginning?
GB: Well, computing has always been my background as I was growing up. I loved playing computer games, but I was very interested in math and science. In fact, up until ninth grade, I thought I might become an actor — I really loved performing and dabbled a bit in philosophy.
What a twist! I had no idea about that, but I plan to figure out the connection between what you're doing now and performing, but let’s move on.
GB: Well, in ninth grade, I felt I had been the star or lead actor in a couple of productions in middle school and high school. I was thinking, I want to double down on something, I feel like I can become the best in some field and really try to push things forward. I felt I could either choose an approach that was more intellectual and hard science oriented or go the arts, performance, and creative route. I ultimately chose the hard science path because I felt that could be the area where I could change the world the most.
Why did you think going that direction would maximize your ability to change the world?
GB: For me, it felt like this — one thing I loved about performance was its communal aspect. I loved improvisation, just creatively thinking about things, bouncing ideas back and forth, but that also required you to be a part of a team that worked well together, and that’s not always guaranteed. What I like more about the intellectual route is it feels like practicing your own skills. But I did learn one thing, that even if you're very good at writing code, that's not enough. In fact, it’s also about bringing in a great team; I think that’s an aspect of my career — really helping to build and shape environments and cultures that can deliver exceptional results.
There’s an interesting aspect, there’s a component about how the environment shaped some of those things. You mentioned you were always the lead in plays and movies. I realize this from when my kids experienced that stage — my daughter had a phase where she really loved musicals — the competition for the lead female role is very fierce, and often just having one competent male who is willing to volunteer means he gets the role every time. Was the competition for the male lead fierce, or were you just unique?
GB: (laughs) I think that might explain it. But let me tell you a story.
How about the reverse? Similarly, in North Dakota, is it not so many people dive into hard sciences, so from that angle does it also seem rarer?
GB: Well, let me tell you two stories. The first is related to performance. My very first paid job was a performing gig. Mannheim Steamroller came to Grand Forks, North Dakota. Have you heard of Mannheim Steamroller?
I have heard of them. Yes, of course.
GB: They were holding a concert and needed extra performers. They needed some people to dress like toy soldiers and walk around for their Christmas holiday concert. I auditioned; at the time, I was a skinny ninth grader surrounded by tall college students.
What they had everyone do was, "Alright, everyone walk in that direction," and then the casting people would compare opinions, then you would have a break, and they would say, "Alright, now walk in the other direction." I noticed during the break, all those college students were chatting with each other, hanging out, while I was thinking about how this job had one requirement, which was to stay focused for six hours during the concert, walking back and forth. So during the break, I would stand there, keep focused, completely immersed in character. In the end, they said, "Alright, we’ll choose this one, that one, and this one. The rest can leave." I wasn’t chosen.
But on our way out, they said, "Actually, we think you’re amazing. We loved seeing how much passion you put into this, so we are going to create a new role for you." So I got a gingerbread man role, and they gave me a full costume. That was my first job. So that’s a bit like trying to break out of the default mindset to get the task done — not always just accepting a role, but trying to find creative ways to get it.
But I think what’s truly great about growing up in North Dakota is that I could be among the very best in the state in the fields I was focused on. I excelled in math. I started taking lots of classes at North Dakota University from tenth grade onward. I participated deeply in math competitions, and then I’d go to national competitions, attend national math camps where I would meet some of the best people in the nation. Those people were amazing; in fact, one of my real honors is that many of the people I greatly admired at those math camps are now at OpenAI. So I had the opportunity to see them in this new field, with new perspectives.
That's wonderful.
GB: But that also means I could really carve out my own path. I began doing math research, and I knew if I went to some of those prestigiou...
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
