Event Replay: Artists and AI: Expanding the Creative Process

[00:00:00] Speaker 1: Hi, everyone, and welcome to the OpenAI Forum. My name is Allie Sandza Wood and I'm delighted to be moderating today's conversation, Artists and AI Expanding the Creative Process. We know that few conversations about AI are as personal or as polarizing as the conversation about creativity. It can quickly become a debate about whether someone is for AI or against it. But we believe that framing leaves out many of the questions artists and creative communities are actually grappling with day to day. Those are questions like, what happens when more people can move from an idea to their art or their artifact with less friction? Can AI give artists more room to experiment and explore? And also, what technical barriers fall and what becomes most important about human contribution? So those are the questions we'll explore today with Eric Zhang, who recently completed his PhD at Boston University and is the lead author of research examining this very question of creative productivity, discovery, and the later market dynamics in the age of AI. Eric's work explores how to move beyond the usual automation story and suggests that AI can widen participation and expand creative agency while also raising important questions, of course, about originality, authorship, trust, and how creative work is valued. So Eric, thank you so much for joining us. Congrats on your PhD. Before we dive into the findings, I'd love to start with just the story behind the work, why you decided to focus on this topic.

[00:01:39] Speaker 2: Yeah. Before that, I want to thank the OpenAI team for putting this event together and inviting me. I just graduated from my PhD, so it's an opportunity of a lifetime. But I really decided to focus on this research topic because, I would say up until the introduction, creativity was, I would call one of the last bastions of the economy in the sense that it's uniquely a human form of intelligence that people said remain untouched by automation. And obviously in the past few years, we've seen that generative AI is really turning that narrative around a bit. And so when you look at it, creativity isn't just about making products. It underpins a lot of major economic activities from innovation to general problem solving. But I think more importantly, it has a very human side to it. So we're talking about self-expression, self-actualization. So when you introduce a technology that can potentially automate parts of that process, the consequences extend beyond what we would think of as sort of basic labor economics or productivity metrics; it is now entering into how we think about socializing around it. And so, you know, from the angle of studying AI, I thought the creative economy was sort of the ideal test bed in my mind, because, you know, understanding this, it's where you can understand the actual strengths and hard limitations of generative models because art is such a highly abstract subjective environment. There's not really clear success criteria that determine what is sort of a viable solution.

[00:03:06] Speaker 1: Yeah, definitely. Which is what I think must be so interesting when you're researching those tension points. So tell us across your research, what do you think audiences should understand about how AI is already changing the process for artists day to day?

[00:03:23] Speaker 2: Yeah, so I would say the biggest shift is going to come down to where artists are spending their cognitive energy. So traditionally the workflow of being a digital artist was heavily weighted towards the actual execution. Bring a concept or idea to life on canvas or on the computer, whatever the case may be. And for the artists, I think now the day-to-day process with AI is looking more like that of a creative director. So what my research shows is that AI can more or less substitute for this initial execution phase. And this sort of makes intuitive sense because the general way that we interact with these models, specifically, we're talking about text image models, you begin with prompting, which you could think of as effectively this human led ideation process, but you're letting the model sort of realize what that idea might potentially look like. So the model is essentially taking care of the visual realization of that idea. And then ultimately it will be the human artist is sort of handling the curation and refinement. And so this changes the workflow in a couple of ways. The first is, like I said, it's sort of enabling this higher volume incremental exploration process because artists are no longer locked into one idea due to whatever costs might be associated with producing it. So they can produce much more content. At the same time, they can effectively explore what I like to call their personal idea space. So you can think of that as like the set of ideas at our disposal that can be combined, reformed in any manner that we see fit. So basically you can see artists experiment with many more ideas.

[00:04:58] Speaker 1: Artists experiment with many more ideas than what was possible before. The second piece to this puzzle is, I would say, at a behavioral level, a potential loss of intentionality and directness when you defer to the model. Because it's so easy to produce so much stuff so quickly, the average novelty of the output will generally drop, even as the best works become more novel, which is sort of what I find in my research. So essentially, that boils down to each piece of output is becoming more marginal, so they're producing a lot more noise, but they're still finding something great at the end. This may be contributing to what the general public might be calling this AI slop narrative that's entering our feed. So it's something to be mindful of. But at the end of the day, the necessary skills, I would say, have shifted from technical execution to now exercising oversight and ideation and filtering. The ability to steer the model through this massive space of possibilities and really tapping into your own sort of ideas in a meaningful way and curating those outputs that are truly meaningful and satisfy some personal criteria for success.

[00:06:08] Speaker 2: So it sounds like to you, or maybe you found in your research, is it expanding the agency of the actual artist or author or director? Or how does that work in comparison to maybe the time they would have spent on ideation?

[00:06:17] Speaker 1: Yeah, so I would say it really depends on how you use it. I think the optimal use case for AI would, of course, be expanding agency because you are not necessarily constrained by, you know, you feel like if I pursue this experimental or niche idea, then I'm sort of without any expected payoff, then sort of I'm just wasting my time. But here, I would say the agency comes from the ability to dictate sort of where the model is headed. I think there's a lot of additional frictions that these artists can add when interacting with AI models to introduce even more agency into that AI-assisted creative process. And so, you know, I think the labor time decreasing is sort of a reality that we can accept. It's sort of a universal thing that we see across AI applications. I think in this case, it's allowing people to express more ideas and for the viewers to be exposed to more ideas, which I think all in all is sort of a net positive.

[00:07:26] Speaker 2: Yeah, it sounds like, I mean, this is what you're describing, is the cost of production is going down. There's less time or cost associated because you can experiment a little bit more perhaps. What do you think that means for people who have ideas, but haven't traditionally had access to either professional creative tools or just don't think that they are an artist? What does that mean?

[00:07:48] Speaker 1: I think for individuals who have traditionally been absent from creative markets for whatever reason, you could say it's lack of background or training, maybe the time passed and now they have families and a full-time job, they can't necessarily put in the time to really hone that craft. Technology, of course, lowers the barrier to entry for these individuals such that they can actually participate. I think before in the creative space, the barrier to entry wasn't due to a lack of ideas. A lot of people have ideas that they want to share. Instead, it was sort of the human capital investment, right? The technical training that goes into it, the sheer time required to actually bring that idea to life, where some people missed that opportunity at some point earlier in their lives. But the research shows that when you remove that bottleneck, we saw this acceleration of novel ideas coming from a more diverse group of creators. This is especially true as communities discover best practices, new model generations are released, and they better understand how to incorporate different models and different features to accommodate their own respective capabilities into their own process. Because it sort of expands the variety of what's being produced, I think it also naturally prompts the broader creative economy to sort of reorganize itself around these different types of values.

[00:09:24] Speaker 2: And so, that's sort of getting more into thinking about competition in the creative economy more broadly because we also have to accept that not everyone uses AI tools, right? There's a lot of traditionalists in this space, whether it be by personal choice, some resource constraints, whatever the case may be. In fact, in my research, I found that only about 7% of creators that I study among hobbyists and industry or communities use any AI tool. And so, there's a question about what happens to the other 93%. And I think there's opportunities there for them as well. So, one way of looking at it is through an economic lens through market segmentation. So, anecdotally...

[00:09:56] Speaker 1: Anecdotally, I think we're kind of seeing early stages of this premium or luxury segment forming specifically around human-made art, and a separate market for maybe more commercially oriented AI media. So I think there's opportunities for both. This isn't really a new phenomenon, I would say, because whenever technology transforms something into a commodity, the traditional labor-intensive means of production doesn't necessarily disappear. Instead, it sort of redefines, and the value is redefined in some sense. The traditional view of craftsmanship, right, the artistic value derived from the process, goes from being an expectation for some people to now being a premium feature. As a result, there is this opportunity where I'd say human proof or human provenance, which is sort of the verifiable history of a piece, comes into play and can be a competitive advantage for both traditional artists and people who use AI in their process.

[00:10:53] That actually leads to another question I had, which was, if AI is making this production easier, what becomes most important in creative work? How do we think about the role of taste, craft point of view, authorship, etc., when we consider the value of the work if the technical barrier is lower?

[00:11:17] Speaker 2: So when the technical barrier to execution is minimized in the way that we've seen, I would say the economic premium shifts focus from the visual aspects to personal taste curation. AI is great at getting you from zero to 75% complete, but in a lot of cases, if we're talking about commercial use cases, a human will still have to jump in, make refinements, and make manual edits to really go the final 25% of the way to make it commercially viable or abide by industry standards. We can see this clearly in our metrics around novelty in the research; we've tracked a consistent reduction in visual novelty, which tells us that the technical craft of producing an image or media is being standardized by models in terms of how we visually consume it. If everyone has access to a tool that can rapidly generate reasonably high-quality visual output, the rendering itself becomes less of a competitive advantage. The baseline for visual quality is raised but homogenized. Instead, what separates successful work is going to come down to the idea being expressed by the individual and the extent to which they refine it, adding their own touch or trying to convey some broader meaning by aligning with cultural symbols and things like that.

[00:13:07] There is the potential pitfall that, because AI allows for high volume generation, there is a dilution of content as the market floods with more average incremental ideas. However, there are diamonds in the rough to be found, and the artist's taste is the crucial economic filter. Meaningful authorship is shifting from how an artist represents an idea to now focusing on what ideas they're representing. Successful AI users are those who possess distinct filters, have developed a voice and a taste, and are able to filter out the noise generated by the model. The ultimate value capture remains firmly with the human capable of closing the gap between an AI output as an intermediate step and refining that into a precise final product that meets personal, industry, or commercial constraints.

[00:13:56] Speaker 1: As we think about creative work and the future of work in general, what kinds of creators or creative roles are changing or maybe newly emerging with the entrance of AI into creative work?

[00:14:12] Speaker 2: The data and anecdotal observations point to the reality that we're now facing. We're seeing a traditional illustrator transitioning into more of a managerial creative director type role. In many ways, this mirrors exactly what we are seeing across the broader economy—not just in creative work but in knowledge work as well. When someone uses AI to automate rote tasks, such as coding or data processing, their day-to-day role elevates from task execution—which may include a lot of mundane things they don't want to do—to process orchestration. They become directors in their own right.

[00:14:54] Speaker 1: The creating economy is experiencing this same structural shift with people using AI tools. With AI, you're effectively giving everyone the same baseline visual production capabilities, so they have to oversee a lot of these other aspects of the process to make that output meaningful. I think in terms of new roles or new skills that might be in demand due to AI, if you actually look at these enthusiast communities for people using these AI tools or developing new pipelines, some of them are doing very different things than just using standard tools out of the box to produce media. Instead, we’re talking about using these tools for highly curated creation and editing in ways that might mirror Photoshop or any sort of animation process like masking out and redrawing a specific detail or shading.

I know some big blockbuster movies have been using these types of techniques, like adjusting lighting and things like that. What’s even more interesting is when individuals using AI tools are taking a still image or even just a simple text prompt, turning those into full-blown animations, mapping out complex storyboards, layering visual effects over a video that was produced just from a simple text prompt. These pipelines are finding their ways into both marketing operations and animation studios. It can be a very overly complex process. It’s not just about typing prompts; it’s about having the traditional aesthetic taste to know exactly what the piece needs, paired with technical capabilities to design and massage a combination of models and features into doing it. Really, they’re treating AI as just one tool within a much larger multi-step pipeline to create something highly specific and meaningful. I imagine this is going to be the new type of creator that’s in demand in the industry, especially in commercial settings—one that combines the technical understanding of these tools with creative domain expertise to come up with clever processes of creating new media.

[00:17:18] Speaker 2: Yeah. It sounds like the baseline that doesn't go away is the creative voice and the creativity of the human mind. The tools are what are changing. Is that how you see it?

[00:17:31] Speaker 1: Yeah, I think that’s the right way of thinking about it. I would say another piece we’re considering is that because AI is lowering barriers to entry, it means people without training can now participate. But there's a paradox here; you sort of need to go through a bit of struggle to figure out what your voice is. If you are circumventing that step of developing the filter and aesthetic preference, we might end up with people being very indiscriminate about what they’re posting and making available to the public in the media they’re producing, which people might not find valuable. There is value in getting your hands dirty, learning, and developing that way. We can call it tacit knowledge or an intangible understanding of how to do things. I think that’s very relevant in the creative world now, and it will only become more valuable as we try to navigate how we attribute value to AI-generated media when people are just typing things. It’s more than that, and people need to take that to heart.

[00:18:54] Speaker 2: I was going to ask you about your research. What did you find in terms of what makes art human, and what part of art needs that humanity behind it?

[00:19:05] Speaker 1: Yeah, it’s really going to come down to your ability to express something. It’s still human expression, right? At the end of the day, anybody can type in a prompt, and the model will spit out something that’s reasonably appealing. It abides by the standard rules of how we prefer to consume visual content. But, you know, art is very much a cultural object; it holds meaning in the sense that it’s about expressing an idea, a story, an experience.

[00:19:52] Speaker 1: And so the more that we can stay true to that framing of how we approach generating media with AI tools, I think the more fruitful we can sort of incorporate this into more of like mainstream operations and how even animation studios or movie studios can start driving more value and creating more value for the eventual consumers of whatever they're producing.

[00:20:22] Speaker 2: Okay, I love that. So I was gonna ask you, creator in your research, creators do the ideation, the curation, the refinement. What do you think would help create a better dialogue perhaps between the places like OpenAI that are creating these LLMs and the creative community? What does that look like in practice?

[00:20:52] Speaker 1: Yeah, that's a good question. I kinda wanna start by even just thinking about how humans are engaging with AI tools because I think, this might sound silly, but I think a lot of the unhealthy narratives that we see floating around can be attributed to human behaviors, right? In some sense, it's users needing to be more intentional, self-aware, and holding themselves accountable when we use these AI tools. And sort of the knowledge work analog that everyone says is like critical thinking, right? We need critical thinking skills. So, this narrative about AI slop that we're seeing floating around, it's a human dilemma. And it happens because some users defer to the model to make all the creative decisions. If you type in a very sort of like loose prompt and let the model take and make all the creative liberties, the model's gonna default to something that's generic, it's recognizably AI, it's producing something that's technically pretty reasonable, but we've seen it before, right? We really can't get much more value out of it. And that's ignoring the fact that some people are also generating low effort content for farming engagement, right? So, I would say the artists who stand out will be the ones who are very self-aware about what they're trying to communicate and how AI affects their ability to communicate that idea, because it's very easy to become deferential to AI, even outside of creative tasks, right? I can speak to it and talk about these AI tools in knowledge work, or even just talking to people in firms that use AI for marketing purposes, right? So, when a model's too good at what it does, it's easy to accept every answer as a plausible solution. So you really need to hold yourself accountable, right? Because convenience can, I would say, pretty easily replace rigor, even if we don't mean for it to. So, that's like one piece.

[00:23:00] Speaker 1: I think when we think about this relationship between creative communities and AI firms, I think it's important to also just reflect on what that ecosystem of how technology is developed and rolled out, what that historically has looked like. Traditionally, we think about the hard sciences, engineer some solution, relatively independent from end users or the economic and policy side of things. They released a new technology, but then they let society figure out what to do with it. And then the social scientists like myself come in afterwards to study the impacts, right? And then the problem-solving step comes in where governing bodies get involved, right? And I would say this way of shaping sustainable policy, it's slow, it's costly, it's highly combative. And there's also just a lack of clarity for every party involved, right? Because we won't see a meaningfully fleshed out framework for another few years. So firms and artists are both operating in a different policy environment than what they will operate with in the future. So everyone's sort of left to figure it out on their own. I think that's like a very classic policy dilemma.

[00:24:10] Speaker 1: So technology is new, impacts are hard to predict. And by the time the impacts are fully understood, the technology is deeply entrenched and it's hard to change things. So I think a healthier dialogue will first start with something like simply aligning on economic incentives beforehand. So I would say right now, we're sort of in this transition period where the economic incentives are quite frankly pretty messy. And so the baseline we have to accept is that general AI is here to stay in the economy and it will naturally permeate many areas of the economy that maybe are unexpected in some ways. But...

[00:24:50] Speaker 1: In some ways, but it's also true that a healthy economy requires sustainable human participation, especially in the arts where it's not just purely economics driving production. So right now I think there's this tendency to take a defensive position from both sides, and I think a healthier dialogue will begin with an upfront, open discussion and willingness to compromise from both tech firms and artist communities to find sort of a workable middle ground. This might look like some economic mechanisms or signaling mechanisms where technology can still scale, and the artists competing in the same market can still build sustainable careers. You know, I think that's like a very high-level lofty goal. I can't speak to specifics because I don't know what's going on on the other side specifically, but, you know, I think in practice it might look like AI firms bringing in creatives into product management goals, for example.

[00:25:52] You want to bridge the translation gap between both parties, and right now an engineer might evaluate one criterion for what determines a model to be suitable for release, but an artist would care more about durability, to what extent, you know, do they have, can they exert agency in that process? So, we need those perspectives when it comes to developing new technology and rolling it out in an economically healthy way. I like to sort of close on a lofty goal that's good, and we are getting a lot of questions about copyright questions and IP. I think we're going to do another forum focused all on that, so I don't want anyone to think we're ignoring that topic.

[00:26:36] Eric, do you want to tell us a little bit about what your next research is about? Give us a preview.

[00:26:45] Speaker 2: Yeah, yeah. So far, I've been talking a lot about what it means for an individual artist to adopt AI tools. I'm looking at how AI impacts individuals' productivity, the novelty of their work, and their workflow. What I'm currently focused on is really projecting that into what happens to markets. I want to understand what happens when we place these AI-assisted individuals into a labor market where everyone's competing for these industry art jobs. Specifically, I'm thinking about some of the structural tensions that are evolving in digital art markets right now.

[00:27:18] We talked about copyright IP and roughly speaking, we also think about the incentive to make your work publicly visible in order to compete to be hired, versus the fear that making it visible also makes it available for model training. So, we're sort of like the middle ground there. I know it's something that the OpenAI team is actively thinking about, and so we're actively tracking how different cohorts—AI users, AI rejecters, established creators—perform over time, and to what extent the production behaviors are reactionary to AI disruption. I'm also thinking about what this means for new creators entering the market.

[00:28:00] How do they strategically position themselves when they enter the markets? In this context, we're not just looking at productivity for the sake of efficiency; we're thinking about productivity as a mechanism for capturing visibility or market prominence. So what's really interesting is where my work is heading is mapping where these creators concentrate their production. We're observing this commoditization effect of producing in genres where you'd believe that AI is most usable. My current focus is on understanding how all these market forces are causing AI-specific creators and human creators alike to adapt. Beyond that, I'm also thinking about where traditional policy might fall short, where it may need to be redefined when it comes to governing the creative economy, and also just trying to quantify what some of the explosive economic impacts of AI on artists are. So that's what you can expect out of me over the next few years. More to come.

[00:28:54] Speaker 1: You'll be busy.

[00:28:56] Speaker 2: More to come, yep.

[00:28:58] Speaker 1: I want to bring in some of our questions that Caitlin has sent me from the community. The first is from Lana Romanova, a frequent forum attendee. Lana asks, could AI shift creativity from producing answers to designing better questions and experiments? Where do you stand on that, Eric?

[00:29:17] Speaker 2: Yeah, I think that's most certainly true. I would say AI exhibits a different form of intelligence than what humans are traditionally familiar with. In the sense, I think of it as like blind spot detection warning for us, right? We don't know what we don't know, and so we pose initial queries much in the same way as I've described. You use AI for ideation or generating reference material for artists. I think it's also giving us more insight.

[00:29:48] Speaker 1: Also giving us more insight into what are other possibilities out there so that we're covering all bases. I think that just can potentially add more value to what we do because it gives us, I'll say, access to the same universal set of ideas. Being able to tap into that can only make us better understand how we can differentiate ourselves, right? So, I think there's a lot of value in using AI to figure out what the right questions are to ask. But again, I think everyone will say critical thinking skills. That's still key here. Don't be deferential, right? Don't just accept what you get at first glance. You're still the operator. You're still the one who needs to go from point A to point Z. So you need to make sure that you steer that with intention. Definitely.

[00:30:52] Speaker 1: Daniel Green, another common forum attendee, thank you, Daniel, asks, while you were in your PhD program, what range of views on AI did you see among creative and academic communities, and what surprised you most? I'll start with the creative communities. I think that one's pretty straightforward. If you browse Reddit, you'll see pro-AI, anti-AI, and people are going crazy over there. You see everything in between, right? Those are the two very extreme perspectives: AI should be banned versus AI has helped me realize self-actualization to an extent I've never been able to do before. The reality is it falls somewhere in the middle. AI is a tool, and that is sort of the general consensus. When we think about the creative space, AI can get you from 0 to 75% complete near instantaneously. But for that to be economically meaningful, socially meaningful, commercially viable, right, it's going to be the human who puts in all the final details to ensure that the message they want to convey is true to what they intended and meets industry standards.

[00:32:34] Speaker 1: In the academic communities, people are all for it. There's so much mundane stuff in research activities that we don't want to do. I think this is where the upside for AI really is—it is not to displace us from our work, but to handle the mundane tasks that we spend hours doing manually. This allows us to start thinking about what is the next big thing, what is the next big question that we can answer so we can broaden and deepen our impact on helping people understand what's really going on in this crazy AI world we live in. In the academic world, it's very much an enabler, giving people more agency to determine where and what they want to study that's most important to them. Our colleagues in the academy and the OpenAI Academy always say, when teaching prompting and other things, "What is the thing you hate most about your job? What do you want to get rid of?" I think that's highly relatable.

[00:33:56] Speaker 2: Okay, our next question is from Brendan Donnelly. Do you think design can lose meaning and storytelling when decisions are made in response to execution rather than being guided by research? I think this kind of gets at what you were saying about the process being part of the art.

[00:34:10] Speaker 1: Yeah, I think, to be quite honest, it's pretty easy to get lost. It's pretty easy for the human voice and sort of whatever they initially intended to get lost. That's why I think the most valuable addition AI can bring to the creative world is the discovery of some of these really complex pipelines that still give the artist agency to determine and refine at every step of the way what will eventually make it.

[00:34:46] Speaker 1: will eventually make it into the final piece. And so there's a lot of communities out there that sort of are working on this, they share their workflows and some of them are absolutely impressive and crazy. It's not something that someone with just a technical background or just an artistic background can accomplish. So I think there's a lot of ways to recapture agency and process. I don't necessarily think it has to even be that the AI is executing things; it could even be people using AI for generating reference material, and then they manually execute once they have a sufficient idea of what they want to really put together. So there's like many different ways of approaching it. And again, I just want to emphasize self-awareness is key, right? Critical thinking is key. Your voice is what will be the economic premium that is added to make sure that whatever you're putting out there is wholly your own and that other people can appreciate it. Yeah, your voice is your currency.

[00:36:02] Speaker 2: Okay, our next question comes from Jordan Holtzman. This question is, with respect to the workshop or marginal output issue, how are artistic AI apps harnessing and propagating individualistic creativity and idiosyncrasies?

[00:36:22] Speaker 1: That's actually a good question. I don't think any of the AI models are exactly doing that. Because again, I think we're talking about an inherently human behavioral conversation here, right? It is up to the user to prompt, verbalize, contextualize whatever they're trying to produce, right? I would say, in some sense, there is a natural friction because an idea that a creative has might be so abstract, might be so novel that in the end, you might end up wrestling with a model, trying to get it to align with your vision, which is why I always say, it's not gonna get you from zero to a hundred; it's gonna get you part of the way there. And part of the way there sometimes is good enough because then the more I think the individual can add their own taste and their own voice to it, the more valuable people will see it. And so that's sort of where we're talking about the traditionalist view of craftsmanship in the arts; that story will always hold true. And I think it will always hold an economic premium that's associated with that.

[00:37:43] I would say, don't try not to worry too much about the model itself. I think you need to worry more about what you are doing with the model.

[00:38:01] Speaker 2: This next question from Jason Deluca, I think kind of gets at that too. Jason writes, "I basically let Chattopadhyay do its thing while I loosely iterate with it for ideas during creative brainstorming. How do you Eric, personally balance structure and openness when iterating with AI?"

[00:38:19] Speaker 1: So I actually kind of set like a hard rule for myself, which is I try to form my own idea based on what I read elsewhere, what people are saying. And then I'll sort of try to formulate a very rough sketch of what that might look like. Then sort of I'll go to chat to BT and ask, you know, are there other angles that you think I'm missing? Here's what my thoughts are. Here's what I think is interesting, right? So essentially, your voice and your taste is sort of still essential to it. It's sort of guiding the conversation in a way that is amenable to what you think is interesting and appropriate. Because I think if you're just going to defer to say, you just present your idea and say, oh, what's missing? You're gonna get some pretty generic things, right? So I think you really need to, again, it's like being intentional about it, holding yourself accountable. I think when you defer too early, then that's where you get into this weird middle ground of, oh, is this too much of like an average idea that other people can easily come up with when they ask chat to be team themselves, right? So that's sort of how I approach it. I found that to be incredibly useful and, you know,

[00:39:44] Speaker 1: And, you know, it also keeps you honest, especially for me. I actually use local elements a lot, which are a bit less capable. I actually find that when a model is less capable, you hold yourself more accountable. If you need to think more for yourself and review the outputs and make sure that it understood what you're saying, maybe if you come up with a half-baked idea, the local will give you something wild, and it's almost like a reality check right in its own way. So, you know, I think, again, your ideas are your own; you should hold on to that. You should respect that and you should use it just as a way of checking some blind spots that you think are most relevant.

[00:40:32] Speaker 2: I love that. I think that's a great place to end on. Eric, thank you so much for your time. Congratulations again on your PhD. Huge, huge accomplishment! Before we go, I just want to tell everyone about a forum we’re hosting next week on July 14th at 1:15 PM, Eastern time, sorry. It’s "Inside OpenAI: How OpenAI teams use Codex to do more." We’ll be joined by our Chief Economist Rani Chatterjee to go through a study that he just did about how teams are using Codex for knowledge work. And then we’ll also have two demos from our OpenAI colleagues, one in design and one on the data science staff. So that should be really cool and interesting. And that is next week. So thank you all again for joining us. Have a lovely rest of your week and weekend, and we’ll see you next time.