OpenAI's $50M People-First AI Fund: How Nonprofits Win

Speaker A: Welcome back to another episode of the Weekly Blitz. We got a little bit of a special guest today, a guy who I met through this podcast, who's now my boss and my co worker and the chief operating officer of Citizen AI, Oliver Belanger. Welcome, Oliver.

Speaker B: Hey Jason. Happy to be here.

Speaker A: Thank you. Julie needed a little family time and I thought, man, let's just dive into what AI companies are trying to do to raise that 28% approval rating that they have with the uh, general public, uh, one of which is awesome for us at Citizen AI and all of which are probably great uh, for society as a whole. So do you want to dive into the OpenAI foundation and what they've released and then we'll touch a little bit on what Anthropic's doing through their, uh, through their nonprofit core program.

Speaker B: Yeah, happy to. So OpenAI foundation has released, uh, a open call for grant applications, uh, for a total of $50 million. They're um, allowing people to ask for up to 10% of their operating budget. Uh, now they're specifically targeting small organizations, highly trusted small organizations in the uh, 500k to 10 million um, dollars dollars, uh, a year in operating budget. Uh, and you know, what it does seem, Jason, that they're doing here is uh, trying to uh, support and help these nonprofits build capacity, understand AI and have the funds to be able to both get used to and practice using the tools as well as, and get learning and development from that as well as, um, purchase uh, purpose driven tools or use different AI tools that are a little bit more on rails than the general purpose tools. And it seems what they're hoping for is for folks to associate, uh, OpenAI and their foundation with um, this forward progress and the use of the technology, uh, trying to boost that approval rating a little bit.

Speaker A: Well, and that's awesome just because this is for lay people. I'm going to dive a little bit into the history of where all this comes from and then let you come back to kind of how this applies to Citizen AI and what the three buckets are in the grant. If you go all the way back to the transformer paper, uh, or the attention is all you need paper and the invention of the transformer architecture that has really the stuff that makes what we understand today as AI work. Because the term AI comes from the 1956 Dartmouth Conference. And so it's not. AI is not a new thing. It's been amongst us. We've had machine learning and a whole bunch of different kinds of AI that I can't Remember the names of. Right. But this form of AI, this large language models really got on the radar in the early 2017s and Elon Musk was kind uh, of solicited by some people to say, hey, Google kind of has this technology. This technology could be like game changing, maybe existential. Elon got a little fired up and threw a whole bunch of money into bringing together some people from the AI research world and the startup world, that being Ilya Sketz, Giver, Dario Amade, Sam Altman, Mir Morati, all these big minds to create the OpenAI, which initially was a nonprofit and its mission was to create artificial general intelligence to benefit humanity. And my perspective is, well, eventually Elon left because there was a little battle over whether him or Sam Altman should run the company. And Elon doesn't like to be second chair. So he got a little upset about it and took off and did his own thing. But the move from nonprofit to for profit, which we've seen in the last year, uh, really was born in emails between Elon and some of these people way back then when they realized that yes, you could really scale these models if you threw a ton of compute and power at them. That's the layest way I can say that. But that is super duper expensive. And so it wasn't really going to be feasible to maintain a nonprofit status and do that. So they became a public benefit corporation this last year. And when they did that, they didn't, they didn't stop their mission of creating AGI to benefit humanity. They just moved it over to a nonprofit that has like $120 billion in this coffer. And so now if you look at what's going on today, leave the data centers out like I think in order to fundraise. And Oliver, you can feedback on your opinion on this, but I think in order to fundraise like we are most driven by our fear factor. It just in our biology. So a lot of these companies have made huge claims about existential risks, government risk, security risk, encryption risks, taking all these jobs risk, so that they would catch people's attention and people would throw money at them to build this stuff without considering the fact that the general public is now like, oh, this is horrible and not something that's going to be good for us. I mean there's very little attention that the general public has on the protein folding things and uh, the efforts around science that uh, Google, DeepMind has had or some of the benefits to humanity. And so now they've kind of flipped that script and said, oh, we really need to change public opinion. We need to show the general public how powerful this technology is in empowering their lives and making their businesses better. And so the way that OpenAI is doing that is by putting out a grant that is targeted at social services, arts and culture, and public information. And I'll stop there and let you take over. And if you disagree with me on any of that, feel free to weigh in.

Speaker B: Yeah, I think it's a good summary. I think, um,

Speaker A: uh,

Speaker B: I think one of the big considerations for both Anthropic and OpenAI to just add a little bit more color is for a very long time, the, The. The folks they cared most about were their investors, and their investors were effectively billionaires or people representing billionaires. Um, and so they didn't really have the incentive to, um, care about, uh, people all that much.

Speaker A: Um,

Speaker B: and the fear narrative, uh, worked, you know, pretty well because that group of folks, uh, that they were targeting for money were already, uh, in belief of the potential risks surrounding runaway, uh, AI and artificial intelligence, uh, having negative impacts. So they kind of tapped into this existing narrative, um, and their argument was, but don't worry, we'll be the safe ones. It could do all these bad things, but if you fund us, we'll make sure they don't happen. Um, and that's what OpenAI did through the nonprofit. And then. And then Dario left OpenAI to do anthropic, and his pitch was, basically, they're not doing it safe enough. I'll do it safer. Um, and so that got built into their. Into their, you know, kind of DNA from, From the early days, as they're now transitioning to becoming public companies, uh, when you believe that the average person are now their investors, never uh, mind their customers, but they're average. They're. They're. They're investors. And so I think part of the reason they're shifting the tone and all of these different things is they're coming to realize that their new class of capital they're catering to is pension funds, is retail investors, um, is hedge funds who are making their decisions based on what retail investors and pension funds do. Um, and so that just, you know, kind of changes. Uh, the. The. What they're kind of going for. Uh, so. So, you know, just as a little bit more of extra color there, uh, just to dive back into the grant. Right. There are three categories, uh, of organizations they're looking to fund. Um, they name them as community support services. Do you provide direct services to people in need uh, community arts and cultural organizations. Are you supporting the arts of your community or the culture of your community in some way? Uh, and community journalism and media, are you providing and unlocking access to information for your community, um, and serving them in that way? Um, this is a really great opportunity for all. And any uh, a nonprofit that fits those eligibility criteria, uh, they're going to get a lot of applications. That's the nature of this. But it is a pretty low lift, uh, and worth mentioning. Ah, as citizen AI, we are supporting many nonprofits with their applications and we are supporting many nonprofits, um, uh, in that process. And so if you are a nonprofit, um, who fits these sort of eligibilities and categories, um, between you know, 500k and $10 million a year in operating budget revenue, shoot us a message. We're, we're happy to help.

Speaker A: Yeah. And the crazy thing about this grant that's so hard for people to wrap their minds around because I worked as a development director. You've worked with nonprofits. We've, I don't write grants. I've written a couple uh, in my, in, in my life. But I always found somebody who was good at that and enjoyed it to do it because usually it is a really daunting process. Now as a development director I have reported back on grants which is also a very daunting process. This is so, not that it is such a simple application. It's unrestricted funds. And the reporting appears uh, to be next to nil as well. Before I pass it over to you, I'll just use an example. There's a platform called Workboard that has um, basically an enterprise level, okr based Monday.com type of platform. Uh, they're out of Silicon Valley and I reached out to them when I was working for a nonprofit and said hey, we're doing this leadership training. Would you, would you sponsor us and give everyone in this leadership training access to Workboard? And they were more than happy to. And all they wanted me to do was join an all hands zoom meeting twice and tell their employees how their use uh, of their platform was impacting business development in, in the nonprofit in Indiana. They were just thrilled by that. Now nobody used it and it uh, like adoption curve for people who aren't techie to use. Techie stuff was a learning curve for me. But, but I think that's kind of what we're looking at here, don't you?

Speaker B: Yeah, the reporting requirements seem to, to be very light. Um, the, the fact that it's unrestricted funds can't Be understated. Uh, it truly is. It goes on your books as unrestricted funds. Um, of course the intention is for it to be spent, uh, on AI projects and processes and the general transformation from, you know, moving you up the adoption curve of AI. Um, however, uh, they are very, very clear in the grant that it is unrestricted funds to be used to serve your mission best. I, uh, think that's because they want to fund a lot of folks. Right? Um, and they do not want to do a lot of labor surrounding funding a lot of folks. Uh, and so I think that's, that's kind of what their motivation is in making it very light. They also probably recognize that at uh, least everyone we speak to in the nonprofit sector has a lot going on. Folks are really busy, um, really focused on their mission and doing the best by their communities that they can. Uh, and they don't have, you know, a whole bunch of time to, uh, jump through hoops. So this is, you know, in terms of scale of hoop jumping and complexity and const. And restriction. Uh, this is on the low end of hoop jumping and it's on the low end of restriction on cash. It's on the low end of everything. Um, and you've even got partners like us who are, who are raising their hand, ready to help and jump through the hoops for you. So if that's of any interest to anyone, happy to help. It is a really, really good opportunity. Uh, they are October decisions or, uh, decisions to be handed out no later than October. Uh, and I suspect, you know, uh, checks will be uh, sent out sometime between October and December 31, 20, uh, 26.

Speaker A: Yeah. Amazing. And they don't even demand. They use their technology which is, which is kind of obscure as well. So if you're a Microsoft shop and you want to invest in training for copilot, um, my friends Brian, uh, Beck and Cullen software are really good with that. But I will say Oliver, like for a while, for about 18 months I was kind of just Claude pilled. I have been using codecs more. I like the fact that it's a natural shift chat interface and it has fixed some things I. It has fixed some things that I couldn't figure out through Claude Cowork. Uh, the different models have different strengths. I'll just say that I'm not, I'm not going to say one is better than the other. But for a while I would have said the. For business, the only one I would use is Claude. I think that Chad GPT has definitely made a shift towards Being less sycophantic and more helpful and it's amazing at design and images.

Speaker B: Yeah, um, they're, you know, they're making progress back. Anthropic has still you know taken the lead, but they've taken a lead through enterprise. Um, and um, you know I think ChatGPT and OpenAI are just looking for other angles and other opportunities to be able to balance uh, the scales, so to speak. Um, you know, while the grant does not require by any means the use of ChatGPT, uh, it probably doesn't hurt to put in your application.

Speaker A: You think?

Speaker B: You know,

Speaker A: Grok, as my uh, primary go to

Speaker B: your arch nemesis is who we plan to work with. Uh, I think actually though this is a really interesting, we're just overall in a really interesting point in um, in the progression of this technology as ah, you know, open um, as Anthropic is in the process of uh, crossing the threshold of profitability which is super interesting. Um, and Both code, both ChatGPT 5.6 and um, Mephos, uh, slash Fable have both been restricted by the US government. Um, uh, though I think 5, uh, point 6 was voluntary, uh, was semi voluntary. And so we're in this really interesting moment where now the latest models um, are being given to you know, Fortune 50 companies, uh, exclusively for a period of time uh, before consumers can touch them, before the rest of the economy can touch them. Which uh, is a very interesting development I think.

Speaker A: Yeah, I don't disagree. It's way over the heads of probably a lot of people that listen to this and that's not their fault. It's way over a lot of people's heads. But I do think LLM routers and training your own on prem open source model is probably going to be uh, direction that we see people going, going forward in a lot of businesses so that they have something to a, reduce their inference cost and B, like you really need to have kind of your own open source base model. Especially when you have something like GLM2 that's, that's extremely powerful so that if the, if so the government decides, oh no, we're going to yank this model that's running your business. You have a backup that, that nobody can touch that and that can hold onto your data.

Speaker B: I think it's more than the government though in this case. And first of all, yeah, I think open source is making huge advances.

Speaker A: Um,

Speaker B: open uh, source will always be behind frontier models. Frontier models, uh, will always be ahead. Um, the thing is that what we're going to get better and better at is supplementing uh, open source models with additional uh, widgets and, and scaffolding and technology to make them more effective. The, the difference in price is just insane. So um, the cost to run anthropic, forget even fable, just anthropic generally is like $25 per million tokens. Um, the cost to run some of the latest open source models is $0.30 per million token. So what that means is open source models bring you back to the price. I mean it basically turns um, LLMs into a similar cost structure to sending text messages. A similar cost structure, uh, a little bit elevated, more elevated to than, than sending um, than than normal compute for websites. But basically it's such a staggering difference in costs that I think for most use cases for most uh, businesses, uh, especially businesses that are looking for uh, integrating into their products. So you know, kind of two core use cases with AI stuff you've got internal workflows and internal things to your business or organization and operations that are really used by your employees to improve uh, internal processes, back office, things like that. Um, then you've got another set of workflows which are integrated into your products which are provided to your customers. Um, in general people have more customers than they have employees. Um, so you know, for serving customers especially where your margin is really, really important, uh, the more you can do with an open source model, the uh, more control you have over what it does, the safer it is from a data and security perspective because you have that control and the better your gross margin because your compute costs are truly four orders of magnitude like they're 1/1,000th or 1/10,000ths. I'm not doing the math off of my head of what using these frontier models is um, is like. And so I think you know, the economics of it is just going to push business towards um, open source because you can't look at two products and sure one is better, but the Frontier labs aren't 10,000 times better than the open source models. They might be a thousand times better, they might be a hundred times better, but they're not 10,000 times better. Um, and so I think really those open source models is where is where a lot of uh, application for the economy generally is going to open up. Um, I think for the internal workflows coding is one and a lot of these different back office functions where your volume is just much lower. Um, I think folks may lean on the frontier models um, to be able to get the latest and greatest and all of that. And this is actually Jason, why I think this shift in how these models are able to be released is very interesting because. And I think it's a strategy, I think it's a communal strategy or a collaborative strategy between government, um, and the labs because I think the labs want this and I think they're just using the government as a cover to be able to do this gracefully and not be shamed in the public markets or in the markets. Um, we're now entering a world where the most expensive, least subsidized and latest and greatest AI models are basically exclusively being shipped out to the companies with the largest balance sheets that can afford, uh, $100 million for like a thousand people to use it. That seems fine. Like they're like, yeah, okay. Um, and so I think that that shift is pretty interesting.

Speaker A: You know, it doesn't have to be a crippler though, right? Because I actually thought about this from the K through 12 implementation stage. What I started thinking about at one point in time was if the schools, the private schools, the schools that have money, are purchasing these higher end uh, vertical AI solutions like school AI, uh, and magic school and these things, like, are those kids going to come out, uh, with a better AI skills than the kids who are in a metropolitan area and just have used to maybe Google free. And what I realized was no, probably the inverse will be true because the kids who have the lower level don't have the state of the art, don't have the scaffolding, are going to have to actually develop skills for working with AI. So some of these bigger companies may get lazy and just rely on how powerful these foundation models are. Whereas some of those who are leveraging AI at ah, not quite the foundation, we are going to have to get crafty and figure out how to use them.

Speaker B: Well, I think so and I think you're touching on something super important which is, um, AI skill fluency, um, and especially for folks who have yet to start their career and are starting their career in the next three to ten years, um, I really do believe we're uh, on the surface, slowly but much faster than it appears, um, entering a world where most of knowledge work is going to be intermediated by AI, is going to have AI in the middle. Um, and that will mean that um, especially as the technology continues to get better, um, even today, it kind of means this already, but especially as it improves, uh, knowing stuff, having what we used to consider, you know, the marketable, differentiated concrete skills won't matter because a person who knows AI not only do, they can not only can they perform that skill? But they can perform a thousand other ones. And so there's a big shift I think in the midst and currently happening, um, that really puts AI in the center in the same way that personal computing and the person and the computer became a like, must have skill for anyone in these information oriented work and roles. There's many jobs that don't require to know how to use a computer. You're an electrician, H vac, plumber, a lot of these great trades that pay quite well, um, and are fantastic pathways.

Speaker A: But look out when recursive self improvement comes to fruition and robotics starts catching up real fast. Sorry man, I had to throw that in there. But you good ahead?

Speaker B: I, I think robotics is gonna, is, is always gonna be a little bit behind because robots are just hard to deal with, um, and building physical things is hard to deal with. And uh, I think Americans are

Speaker A: uh,

Speaker B: less into robots than, than they, they are into AI. Like they're, they're even more fearful of Robocop than they are of um, you know, like Minority Report. Exactly, Minority Report.

Speaker A: There.

Speaker B: We're more fearful of RoboCop than we are of like pre med of uh, predicting future crime and acting upon it. Uh, so I think we'll have some slowdown there. Plus it costs a ton of money. Um, whereas China is like, they like robots, they like robocop.

Speaker A: Um, they do, they do all of Asia I think, or at least Japan and China love it. So let me, let me tell something on what you're saying because I do think there are some new skills. And, and I've never really told you that this is why I do this, but I had this conversation, I was part of a workforce development conversation with Tech Point the other day. We broke up in uh, a little groups and we were talking about the future of work and one of the things that people were saying was like, well, you know, people with an 8th grade education are using AI to send emails that sound like they're more intelligent. And I'm like, who gives a damn? You're focusing on the wrong thing. Like if you have KPIs and they're uh, they're perform, they're delivering on their deliverables. Like who cares if AI helps them communicate better? That is, that is irrelevant. But what the skills that I think they need to focus on, you can weigh in on that. And then I'll throw and then, then we'll bend into what I do think is becoming important.

Speaker B: I, I, I do think we should care. I think we should Care, because it's a great advancement for that population. Yeah. The fact that, the fact that exposure and uh, skill in being able to write in a certain fluency of a certain way is not becoming a barrier for people who have the ability to perform the activities at the same economic level so that they don't get prejudged out of it because the email they sent doesn't sound quite right, um, to the people receiving it, we should care. That's a huge advantage that is better for the economy, that is allowing more economic mobility up. And so I think it's a huge positive.

Speaker A: I agree. And I said that out loud. I was like, listen, nobody in this room knows it, but I probably have a 6th grade math level. I hated math. And yet that does not hold me back. I can use AI to hide that. And it's not an intricate enough part of my job. I'm not a number cruncher or a scientist, so that's actually just a win for me and whoever I'm working for. But the skills, I do think that are, that we, uh, you know, people talk about taste and discernment and those are all, all those soft skills are kind of hard to define. But the ability to create evals and heuristics and rubrics, I think like that is how you optimize your human hybrid workflow. And that's a lot of times I never had any experience with that. I do have some experience with like traditional design, but that's getting a little bit more into like educational and engineering design. And so a lot of times I'll try my hand out and send over to you what I got to see what your thoughts on it are, because I know you think more along those lines. But I think data literacy and fluency and how to set up rubrics and how to evaluate outputs are three super valuable skills going forward.

Speaker B: Yeah, I think it's also hard. Um, there is a, there's a form and substance aspect to those tasks. Um, and, and so knowing the form of the skill is valuable and important, um, because it allows you to basically create the scaffolding for the AI to be able to operate effectively. Um, part of the problem in it though is the substance. And what I mean by that is a rubric, an eval, any of these different processes and mechanisms that are about evaluating the output and the value and effectively the quality of the output, um, require the, uh, require within it to be the understanding and knowledge of what makes the output quality. And it's this really interesting thing where, um, the form can't do that. Meaning just like you can have the perfect process for evals, the perfect process for Rubrik, nailed the skill, know it inside and out, can do it at an Olympic level, um, and without having the prerequisite, um, understanding. This is where we get into taste, discernment, judgment, that pile of things which really just all equivocates to like having an intuition for what good is in this specific domain. Um, you can't actually inject and incorporate into it um, the, the ability to judge or the ability to identify or make the framework, the scaffolding produce quality output. Um, and so very interestingly and, and when you ask, when you ask the AI to do it, um, if you're lucky it will produce like C to B level uh, capability. And that's just because it uh, you know, it averages out the Internet and the Internet on average does C work, um, which is in many cases better than a person who doesn't know how the thing works. Because a person who doesn't know how the thing works does, you know, E work. Um, so it's a movement forward, right? It's a progression. Um, um and so this is where I think you know, to the point you're making Jason, like partnerships between people who have domain expertise, um, and people who know how to equip and set up and use these AI skills are so valuable um, because no one is an expert in everything, no one knows everything. It's, it's not, it's not within our human capabilities yet. Fast forward 20 years. AI isn't everything. Maybe it is but um, and I think uh, like the ability to have people with different expertise in different domains be able to lend their support with each other, to be able to provide that positive signal, that judgment, that discernment and work with another person to encode it into the evals effectively. The category of uh, quality improving scaffolding uh, is critical to be able to provide and use AI to produce economically useful work. Um, especially at the higher levels of quality. Uh, you can just for many emails. Right, so back to that example you mentioned of, of folks with an 8th grade education using it to sound like they have, send an email that sounds like they have a, you know, college education, um, which is a wild point for someone to think of as negative. But um, uh, we're just gonna move on from that. Um, the AI doesn't need scaffolding for that out of the box. It can just do that um, pretty well. Uh, so there's no need. Uh, but when you're Getting into more complicated outputs, when you're getting into a lot of different things, uh, it, it doesn't produce quality output and needs tinkering, needs a little bit of love and care to be able to produce quality output. And I think this is one of the hurdles to kind of two things come to mind here. This is one of the hurdles to wrap it back around that folks using AI for the first time struggle with, because out of the box it doesn't produce quality anything. Um, it produces the average of what the, it produces the average of what Reddit would come up with effectively. And in many use cases that's actually fine. Reddit does great. You know, if I'm looking for, for a restaurant, um, that I want the like not bullshit answer on what's good or what's bad in an area, Reddit is probably the best place to go. Um, uh, if I'm, you know, trying to, uh, come up with a new molecular structure. Ah. Or trying to fold proteins, then perhaps Reddit is not the place that I, that I want to go. Um, um. And, uh, I think this is a place that people struggle with because they'll use it for some things and they're like, this is amazing. It gives me exactly what I need. And then they try to use it for a different kind of task that requires more domain expertise, and they have that domain expertise and then they evaluate it and they go, this is shit, I can't use this. Um, and then the thing that people make, you know, kind of struggle with is they'll use it to do stuff they don't have domain expertise in where they themselves can't evaluate the quality output, and they effectively go, there is output, it works. And then they hand it off to someone, and then someone looks at it that has that domain expertise and goes, this is shit. What are you, what are you sending this to me for? So I think the understanding and ability to work with that and develop that is one of the biggest hurdles people run into. And being able to use AI to develop economically useful, um, uh, uh, processes. Uh, and this is where I think in many cases the partnership of folks who have a good sense of the AI skills and designing AI workflows with domain experts comes in together really well because domain experts are oftentimes domain experts in shit that is not AI. Um, and in many cases want to remain that way and they'll learn and all of that. And I actually think, um, that's a huge opportunity for, uh, folks who are entering the workforce in the next three to ten years. Um, is for this window of time because folks entering the workforce, what they lack is domain expertise. They have energy, they have whatever the hell they learned in school. They have um, whatever level of school they had. They um, have time and they have a, uh, hopefully uh, you know, low lifestyle cost needs. They uh, don't have a family to support hopefully usually, but sometimes they do. Um, and so I think there's a great opportunity in the short, in the, in the immediate term for these folks who have spent the time to learn the AI skills, the scaffolding and all of that, to work really closely with domain experts and through that process and through that trade, um, enable the domain experts to take advantage of the scale of AI, um, while simultaneously learning from the domain experts and absorbing their discernment, their judgment, their understanding of that domain and allowing them, themselves, those junior employees, uh, to be able to then apply that domain to the same level of quality and capability, uh, as those senior employees, um, but along with the AI skills. And so I think this is one of the greatest opportunities in the next five to ten years for folks entering the workforce, um, because people in late career don't want to learn new stuff usually unless they're forced to.

Speaker A: I'm going to 100% agree with you and just rephrase it slightly in the way that my mind kind of thinks of it. So I love this idea of understanding the theory of mind of how large language models operate like uh, because they all do kind of operate slightly differently. But there is this almost psychological understanding of how does this particular model in best input natural language to output what I'm looking for. So if you are, if you, and, and you do not have to be a software engineer, I've gotten pretty damn good at that if I, if I will say, right, so if you understand the theory of mind of these large language models and then can enter a company and pull all this tacit knowledge that's in all of these uh, domain experts out of them and, and use that to, to create workflows and agentic scaffolding inside of AI like you become very valuable. The reason, Oliver, that I think a lot of people don't understand, yes, AI has advanced most in code because it's a bunch of software engineers and that's what they care the most about. But that's not the only reason. The reason is like there are systems and records and steps and like all that stuff is completely laid out. When it comes to the engineering process, any basically engineering process, when you get into white collar tacit knowledge, sales Marketing type of work like that stuff's just not written down everywhere. Hell happy these people don't have like defined KPIs. Like if you can't, if you can't do a slash goal then you better not run an agentic loop or you're just spending a whole bunch of money. Right? So m. Maybe you, maybe you could unpack that just a little bit. But I think that that kind of plays into as well that a lot of this stuff is just in people's heads. It's, and that's it can't be there for a big company or even a medium sized company to leverage AI.

Speaker B: Yeah, just to touch on that. I mean so in. And then part of my background is software, uh, engineering and I got into the tech industry through a workforce development program and all of that. But um, so in the software engineer trade profession the thing you just called out, the tacit knowledge that is in people's heads has a name actually. And so what people call it is tribal knowledge. And um, not only does it have a name but it is a form of technical debt um, that is actively managed uh, has been for 20, 30 years. Um and oftentimes is considered one of the difference like the, the degree to which it is managed is oftentimes considered one of the different markers between a high quality um, engineering team and a low quality engineering team. So the very precise um, uh thing you're, you're mentioning uh, is a core uh, KPI sort of speak north uh, star metric so to speak of, of the culture of this profession. And because of that, you know, if we were to put it on a continuum of fully 100% they're all tribal knowledge is, is on some form of paper and you know, zero. It's all in everybody's heads. Uh engineering culture pulls you towards the center of call it 50% or something like that. Now 50% may not sound like a lot, but it's a hell of a lot better than the norm which is like 10% if that. So uh, it, it's, it's integrated into the way the, that profession operates which is of course made it easier to move forward. It also to your point, the work product is written word and the work product is code. And code is literally structured logic that is mathematically backable. Like it, it doesn't work if it's not logical. Marketing, uh, is written down but it is in many ways arguably illogical, irrational and designed not to do A plus B equals C but to do A plus B equals C squared. Don't worry about it. Um, love it.

Speaker A: Yes. So, right.

Speaker B: So it's, it's inherently harder for, for a learn, you know, a model to be able to, to kind of pick that up. I think that's, uh, everything you're saying, I think is true. I also think there's more of the culture of engineers that make that. There's two angles to it that further amplify the reason that that specific use case has exploded in adoption. And other, the other ones have one. It's software. Engineers and engineers generally cannot help themselves. But tinker. Yeah, that's why they're engineers, because they're like, I wonder if I can make it do this. Like the, the, the prototypical, like, engineering joke or software engineer joke. Uh, it's a great little comic. And, um, it's two scientists interacting with, uh, alien technology that has a screen, and they're going, I wonder if it can cure cancer. I wonder if it can do this. I wonder if it can do that. And then the next panel is an engineer hopping up from behind and going, I got doomed to run on it. And, uh, the culture is very much one of being, like, looking at a system of some sort and asking the question, I wonder if I can make it do this purely for the fact of, like, I, I like, why? What is the economic outcome? What is the, what do you get? You know, what's the external reason, uh, you want to do this? And the engineer is just like, what are you talking about? I, I, I just want to see if I can make it do that. I just want to see if I can tear it apart. Right. I mean, oftentimes engineers, you know, there's stories of kids being like, I took apart the radio or I took apart the tv. Did you put it back together? That's a separate question. Um, so I think the culture of it is one of tinkering, which means the very fiddly nature of LLMs, uh, is not a negative but a positive, because every engineer gets to fiddle with it in just the way they want. Uh, so that's one cultural element, I think, that really pushes it forward. And then the final economic element is if you figure out how to make AI write code. Well, AI is produced by code. So for anthropic brilliant move on their part, it's actually pretty straightforward logic.

Speaker A: Huh?

Speaker B: Huh. If we can figure out how to use AI to write more code, we can then use AI to write better AI and we can accelerate our advancements in relation to other providers because we will always be better at, uh, using our own tools than anyone else in the world. And they are like, it's not even close. They, uh, are the most AI native company in the world. And the, uh, way they operate sounds like science fiction to, you know, most organizations. And so part of it too is if you had $100 billion, wouldn't you put it, you know, wouldn't, wouldn't it be neat to be able to invest that money into a workflow that both the whole world will pay a shit ton of money for, but also you can leverage better than anyone else to compound your core dome your core advantage faster. And so I think this is, this is the other reason that they just invested in it more because they were like, if we figure out code first, we win.

Speaker A: Unless you maximize, uh, Paperclip. But, uh, I think that's the thumbnail is going to be the guy from, uh, from Doom. Oh, that's Duke Nukem going, want to dance? Want to dance to a paperclip? But we digress. Um, we started with the nonprofit space and we're almost out of time and I know we still have a lot of work to do trying to help people leverage this open AI grant. There is another grant I'm going to read real quickly. Anthropic has put out $150 million grant. They're looking, it's called cloud core. Cloud core is a big initiative and it's going to invest in placing 1,000 early career fellows into 400 nonprofit organizations for one year to help organizations adopt AI. Um, so that's a much more, uh, finite way of basically sending four deployed engineers FDEs out into the nonprofit world at only 400 nonprofits. I'm guessing those are going to be probably fairly large. But if that's something you're interested in exploring, we will put a link to that in the show notes. If you want to get a hold of us to explore the OpenAI grant particularly, or in the state of Indiana, because that's really where we're launching, then please reach out to me, uh, reach out to us, go to our website. All that will be in the links. We were going to kind of get into the DeepMind departures. Go ahead.

Speaker B: One more thing on the OpenAI grant, it's a July 15 submission date. So time is of the essence.

Speaker A: Time is that grant is moving at the speed of AI. We were going to get into the DeepMind departures. And the fact that their, uh, Nobel Prize winner just went to Anthropic, they got to share the Nobel Prize with Dario for, for Alpha for creating Alpha full. We're not going to get into all that. I'm just going to ask you, you can say whatever you want, but I'm going to ask you is Dario going to stay or is he going to go?

Speaker B: Sorry, what?

Speaker A: I'm not Dario Demis. Demis. Is uh, he going to go?

Speaker B: I believe Demos is staying at Google, um, because he's the AI lord of Google's resources and why would he, the only reason he would leave is to go be the AI lord of someone else's resources and there are very few piles of money as large as Google's, um, so I doubt that he'll leave Google. I think people are leaving Google. I don't think people are leaving Google.

Speaker A: Jeff Bean just left Google. I mean that's, that's a big guy.

Speaker B: I think that's the wrong way to look at it. I don't think people are leaving Google. I think people are joining in for opic.

Speaker A: Yes.

Speaker B: And I think what you're seeing is not people. I think Google is continuing to do the Google thing. I'm watching every couple of days more integration that's really thoughtfully and well done into the G suite. It's actually coming together quite nicely these days. Recently, um, they launched a, ah, feature that was catch up on a document so you can, you know, you, you get on a shared document, 100 people start fucking around in it and like moving it forward and everything and you're like, you come back to and you're like, what, what happened here? What's going on? They now have a catch up feature that you can press a button for where it will just give you a AI, uh, uh, summary of what's changed uh, since the last time you looked at it, which is a really cool feature. So I, I don't, I think looking at it as people leaving Google is, is not really the, the, the thing that's happening. I think really the story is people are joining Anthropic regardless of where they're at because it's the pull to Anthropic that is unique, special and interesting. And it really just comes down to Anthropic having like had a flawless few years um, of execution. They are just winning, uh, uh, and, and they are winning in by every measure of startup technology. Like any way you count score, they have beaten the, you know, world records across the board. So it's about joining the most AI team in the world, joining the best team for this thing in the world. Less, um, so than it is about leaving the number two or leaving the Number three or whether it's two or three, it's really just Anthropic pulling in, um, the best talent in the world as they have made now a clear lead over everyone else.

Speaker A: And I don't disagree with you at all. I wonder. So one of the big researchers just went to OpenAI as well. So I think if I just put myself in this like we're on the. If I really am a believer in fast takeoff, right? If agentic workflows and harnesses and all that, it's something that I'm really into, I might go join OpenAI because Karpathi is over there now. And uh, that's probably going to be a big deal, right? But if, if I really believe that we're on the cusp of recursive self improvement and that is going to be fast takeoff. Even if I'm Demis established, I have to know that Anthropic is ahead of everybody else on that. And do I want to be, do I want to, do I want to try to race to keep up with them or beat them, or do I want to just go join them and be part of the party? Because once that happens, all bets are off.

Speaker B: I think the thing you're not taking into account there is um, that

Speaker A: uh,

Speaker B: Dario Demis, all of these folks, uh, Elon Musk, were part of the Singularity, um, Institute. They were part of Miri, later became Miriam Mission Intelligence Research Institute. So they, their peers, they were part of the same class, sort of speak of people who saw the future of what AI could be and had the skills and capabilities to bring uh, it to that. And they had different philosophies on, on what it would mean uh, to bring Overlord AI into, into reality. Um, and so the reason Demis wouldn't join Anthropic is because he would have to accept Dario's vision of what it means to create, you know, our next AI overlord. And he doesn't want that. He believes in a different vision of what our AI overlords should look like and is building that. And so I think like the only reason he would leave Google's big pile of resources and I think he just looks at it as Google has a large pile of resources for him to use to try to make that is to control another big pile of resources, um, to achieve that vision. I think the, the biggest players of this game, um, are ideologically driven, um, towards a specific vision that they see. That vision may be very self serving, but it is a specific vision that they see, including la Musk. Um, and those visions are incompatible, and the ones that are compatible are just being folded in. So anthropic is getting a lot of traction. And so the folks that are joining them are just seeing their own vision for. At the highest levels, seeing their own vision for where this technology goes, seeing it in anthropic and throwing their chips in with that and being like, I'm good with this. Um, probably with a little bit of coaxing from, From Dario about how well actually my vision is your vision and blah, blah, blah. Um, but the folks that see themselves as, as peers, as the spear points of these different philosophies inherently are divergent in what they believe it should be. And, and in many cases, these folks do see themselves, um, except for Sam Altman, as safeguards and stewards of bringing true intelligence to, To. Of birthing machine intelligence and human created intelligence. Like they, in many cases, these folks at the very, very top see themselves in that way. Um, and so I think they're not. There's not just economic considerations. Uh, and even the economic considerations in many cases are for the purpose of having the resources to be able to do these ideological driven ambitions.

Speaker A: That is, that is very wise. I mean, you like, there's. Where the hell is. Is Ilya sets giver. Uh, Sarah Fryer should probably be the CEO because she talks a lot better than Sam, and Sam should just be outraged like every time you talk. I just want to talk about 10,000 more things, but I don't want to keep you on forever. So last words. Thanks for joining me, Oliver. Any last words that you have. We'll take the rest of these conversations offline. Most of my audience probably has no idea who the Elliot givers anyway, so, uh, we'll save that one.

Speaker B: I. I think my last words would be, um, there's three money. If you're a nonprofit, between 500k and 10 million. Uh, and if everything we said sounded either overwhelming, kind of boring, or like, I don't want to deal with that. Uh, but you. But you like unrestricted cash. Uh, you should send us a message because we'll help you win the unrestricted cash, uh, further. Anyone who's in their early stages of exploring AI and is trying to get more into it, truly the number one piece of advice I can give you is use it to do shit that, you know, annoys you. I think, Jason, you like this, this framing. But. But use it there. It. It truly is different than anything before. It is nothing like other things. You can only build an intuition of what's possible in the world that is now supplanting our current one. Um, if you have a familiar reality with the technology and you can only develop that by using it. So use it. Use it poorly. Using it poorly is fine. Someone could teach you how to use it better. But start using it, if only to have an understanding of it.

Speaker A: I think that's great advice, I would say as a non technical person. Tinker. I love the tinker. Tinker with stuff, uh, with it on stuff you are good at because then you'll know whether the output's good or not and then quickly figure out what you don't like doing and ask it how to automate that. Ah. Or reduce the burden of it. And you will be very happy with, uh, you may just get your own chatgpt moment. All right, thank you, Oliver. We will do this again.

Speaker B: Bye folks.