The AI Daily Brief: Artificial Intelligence News and Analysis - Andrej Karpathy on How AI Empowers

Starting point is 00:00:00 Today on the AI Daily Brief, former Open AI co-founder Andre Carpathy on how LLMs flipped the script on technological diffusion or, as I'm framing it, AI empowerment. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes. Hello, friends. Quick notes before we dive into today's show. First up, as I've been mentioning a couple times, for those of you who are looking for an ad-free version of the AI Daily Brief, you can now head on over to patreon.com slash AI Daily Brief to find that. Lastly, something I want to gauge people's perspective on. The AI Daily Brief community has been hugely supportive of and important in the superintelligence

Starting point is 00:00:43 story. We're considering reserving part of our current round for investors from this community. However, I'm trying to gauge interest. If this is something you think we should explore, send me a note at NLW at Bsuper.a.i With super in the title. Thanks in advance for your perspective. And with that, let's get into today's show. Hello, friends. Welcome back to another long reads episode of the AI Daily Brief.

Starting point is 00:01:05 One more quick reminder before we get into it. As you are listening to this, I am in Florida, probably just about to surprise our kids with their first Disney World trip, which should be great. As I mentioned, though, this means that next week's episodes are a little bit different. We have a slate of really interesting interviews, lots of talk about agents and vibe coding and big technological changes, stuff that I think will be a really interesting and enlightening change of pace. If some crazy thing happens, you can be assured that I will find a way to get in there and share some content with you. But in the meantime, assuming that that doesn't happen, we will have a week of pre-recorded interviews, and then we will be back at the end of next week

Starting point is 00:01:39 with an Easter long read Sunday and then normal episodes to follow that. For now, though, let's jump over into this recent post from Andre Carpathy. It's basically a blog post but shared to X. And it's always a real treat when we get to read a piece from one of the big thinkers in the space, especially because these types of pieces tend to be more conversational, let's say, than the essays that get published in an op-ed section in Bloomberg or something. By choosing to put this on X, Andre's inviting discourse and conversation, and we've seen in the past what happens when Andre invites discourse and conversation. When the man introduced the term vibe coding, which, although it's kind of been warped from how he was originally using it, is obviously one of the most

Starting point is 00:02:17 influential concepts of the year. In any case, this piece is called Power to the People, how LLMs flip the script on technology diffusion, and we're going to read it first, and then I'll come back and discuss it. Once again, you guys are on a roll. This is actually me reading it. rather than AI. Andre writes, transformative technologies usually follow a top-down diffusion path, originating in government or military contexts, passing through corporations, and eventually reaching individuals. Think electricity, cryptography, computers, flight, the internet, or GPS.

Starting point is 00:02:48 This progression feels intuitive. New and powerful technologies are usually scarce, capital-intensive, and their use requires specialized technical expertise in the early stages. So it strikes me as quite unique and remarkable that LLLL display a dramatic reversal of this pattern. They generate disproportionate benefit for regular people, while their impact is a lot more muted and lagging in corporations and governments. ChatGPT is the fastest growing consumer application in history, with 400 million weekly active users who use it for writing, coding, translation, tutoring, summarization, deep research,

Starting point is 00:03:19 brainstorming, etc. This is not a minor upgrade to what existed before. It's a major multiplier to an individual's power level across a broad range of capabilities, and the barrier to use is incredibly low. The models are cheap, free even, fast, available to anyone on demand behind a URL or even local machine, and they speak anyone's native language, including tone, slang, or emoji. This is insane. As far as I can tell, the average person has never experienced the technological unlock, this dramatic, this fast. Why then are the benefits a lot more muted in the corporate and government realms? I think the first reason is that LLMs offer a very specific profile of capability, that of merely quasi-expert knowledge and performance, but

Starting point is 00:03:59 simultaneously across a very wide variety of domains. In other words, they are simultaneously versatile, but also shallow and fallible. Meanwhile, an organization's unique superpower is the ability to concentrate diverse expertise into a single entity by employing engineers, researchers, analysts, lawyers, marketers, etc. While LLMs can certainly make these experts more efficient individually, e.g. drafting initial legal clauses, generating boilerplate code, etc., the improvement to the organization takes the form of becoming a bit better at the things it could already do. In contrast, an individual will usually only be an expert in at most one thing. So the broad quasi-expertise expertise offered by the LLM fundamentally allows them to do things they couldn't do

Starting point is 00:04:39 before. People can now vibe code apps. They can approach legal documents. They can grach esoteric research papers. They can do data analytics. They can generate multimodal content for branding and marketing. They can do all of this at an adequate capability without involving an additional expert. Second, organizations deal with problems of a lot greater complexity and necessary coordination. Think various integrations, legacy systems, corporate brand or style guides, stringent security protocols, privacy considerations, internationalization, regulatory compliance, and legal risk. There are a lot more variables, a lot more constraints, a lot more considerations, and a lot lower margin for error. It's not so easy to pull all of it into a context window. You can't just vibe code

Starting point is 00:05:19 something. You might be one disastrous hallucination away from losing your job. And third, there is the well-documented inertia of a larger organization, featuring culture, historical precedents, political turf that escalate in periods of rapid change, communication overhead, retraining challenges of a distributed workforce, and good old-fashioned bureaucracy. These are major headwinds when it comes to rapid adoption of a sparkling new, versatile, but shallow, and fallible tool. I don't wish to downplay the impacts of LLMs and corporations or governments. But at least for the moment and an aggregate across society, they have been significantly more life-altering for individuals than they have been for

Starting point is 00:05:53 organizations. Mary, Jim, and Joe's are experiencing the majority of the benefit, not Google or the government of the United States. Looking forward, the continued diffusion of LLMs, of course, depends on continued performance improvement and its capability profile. The benefit distribution overall is particularly interesting to chart and depends heavily on the dynamic range of the performance as a function of capital expenditure. Today, Frontier-grade LLM performance is very accessible and cheap. Beyond this point, you cannot spend a marginal dollar to get better performance, reliability, or autonomy. Money can't buy better chat GPT. Bill Gates talks to GPT4O just like you do, but can this be expected to last? Train time scaling, increasing parameters and data, test time

Starting point is 00:06:32 scaling, increased time, and model ensembles increase batch are forces increasing the dynamic range. On the other hand, model distillation, the ability to train disproportionately powerful small models by training to mimic the big model has been a force decreasing dynamic range. Certainly the moment money can buy dramatically better chat GBT, things change. Large organizations get to concentrate their vast resources to buy more intelligence, and within the category of individual two, the elite may once again split away from the rest of society. Their child will be tutored by GBT8 Pro Max High, yours by GBT6 Mini. But at least at this moment in time, we find ourselves in a unique and unprecedented situation in the history of technology.

Starting point is 00:07:10 If you go back through various sci-fi, you'll see that very few would have predicted the AI revolution would feature this progression. It was supposed to be a top-secret government megabrain project wielded by the generals, not Chatchipit, appearing to, basically overnight and for free on a device already in everyone's pocket. Remember that William Gibson, quote, The future is already here, it's just not evenly distributed. Surprise, the future is already here, and it is shockingly distributed. Power to the people. Personally, I love it.

Starting point is 00:07:37 Today's episode is brought to you by Vanta. Vanta is a trust management platform that helps businesses automate security and compliance, enabling them to demonstrate strong security practices and scale. In today's business landscape, businesses can't just claim security, they have to prove it. Achieving compliance with a framework like SOC2, ISO-2-2, HIPAA, GDPR, and more, is how businesses can demonstrate strong security practices. And we see how much this matters every time we connect enterprises with agent services providers at superintelligent. Many of these compliance frameworks are simply not negotiable for enterprises. The problem is that

Starting point is 00:08:12 navigating security and compliance is time-consuming and complicated. It can take months of work and use up valuable time and resources. Venta makes it easy and faster by automates. compliance across 35-plus frameworks. It gets you audit-ready in weeks instead of months and saves you up to 85% of associated costs. In fact, a recent IDC White Paper found that Vanta customers achieved $535,000 per year in benefits, and the platform pays for itself in just three months. The proof is in the numbers. More than 10,000 global companies trust Vanta, including Atlassian, Kora, and more. For a limited time, listeners get $1,000 off at vanta.com slash NLW. That's V-A-N-T-A-com slash NLW for $1,000 off.

Starting point is 00:08:54 Today's episode is brought to you by Super Intelligent and more specifically Super's Agent Readiness Audits. If you've been listening for a while, you have probably heard me talk about this. But basically the idea of the agent readiness audit is that this is a system that we've created to help you benchmark and map opportunities in your organizations where agents could specifically help you solve your problems, create new opportunities in a way that, again, is completely customized to you. When you do one of these audits, what you're going to do is a voice-based agent interview where

Starting point is 00:09:25 we work with some number of your leadership and employees to map what's going on inside the organization and to figure out where you are in your agent journey. That's going to produce an agent readiness score that comes with a deep set of explanations, strength, weaknesses, key findings, and of course, a set of very specific recommendations that then we have the ability to help you go find the right partners to actually fulfill. So if you are looking for a way to jumpstart your agent strategy, send us an email at agent at B-Supert.aI, and let's get you plugged into the agentic era. So I think there is a ton that's interesting in here.

Starting point is 00:10:01 And I'm going to take it in honestly no particular order, just kind of my set of thoughts as I reflect on this piece. The first is that I think it's actually a very salient point that helps us understand how AI is improving things right now to recognize that corporations are groups of specialists whereas an individual inherently can only be a specialist in so many things. And so given that an individual is a specialist in very few things and a novice in the vast majority of things, AI's capability to take them from novice to intermediate or novice to adequate in terms of design, coding, whatever it is, is a greater change in general

Starting point is 00:10:40 than a person who's just operating in their specialist capacity inside the context of the corporation they're contributing to. Basically, the band of use cases that are valuable to an individual operating as an individual are inherently higher than the band of use cases useful to a specialist operating as a specialist. I think it's a really salient point, and I also think that it will influence how these things take root even inside enterprises and corporations. One of the interesting and non-obvious outcomes, although it feels like it's getting perhaps more obvious now, is a stronger differentiation within side companies between core

Starting point is 00:11:17 functions and secondary functions. And when it comes to those secondary functions, more possibility that the capabilities of a novice who's a specialist at the main thing using AI tools to do those things might be good enough. In other words, companies might decide not to hire certain skill sets and expertise that they would have before because it's not core to the business even if it's something that they do, yes, have to do. And the AI-ified version is good enough. Now, obviously, if you're sitting there screaming agents, but agents. I think you're correct to recognize that this seems especially where we might see the rise of people opting to hire digital employees instead of human employees. One of the ways to embody and embrace that good enoughness is to pick the low-cost

Starting point is 00:12:02 agentic option that, yes, maybe still can't compete with best-in-class human versions, but for a company for whom that thing is important but not mission-critical, that's totally sufficient. One of the themes that you'll hear a lot next week in our interview conversations is the idea that a lot of where agents are going to start is going to be on the margins. In sales organizations, in marketing groups, not that those things aren't important, but they are separate from the core function of whatever the business is. A second thing, though, that I think is important to note is while I agree in general that if you just take an individual operating in their own capacity and an individual operating

Starting point is 00:12:37 in their corporate capacity as a specialist, AI and LLM specifically have, have had or have the potential to have more broader impact on that individual operating in their own capacity, I think that Andre's argument perhaps might mislead people in terms of understanding how big an impact and how fast these technologies are already having in the enterprise sector. If you look at the previous rate of change and rate of adoption of technologies in the corporate sector and compare it to AI, there is absolutely nothing comparable. The speed with which enterprises and businesses have radically reoriented themselves to at least attempt to adopt these technologies is totally unlike anything we've ever seen. There is an understanding up and down the

Starting point is 00:13:20 organization that these are hugely disruptive forces. They go beyond just new tooling. They are unlocking totally new types of efficiencies as well as totally new types of opportunities, which we have barely scratched the surface of. And relative to the pressure of general corporate inertia, the adoption is actually happening incredibly quickly. Now, like I said, that doesn't undermine his point that there's something really powerful about the fact that people are figuring this stuff out on their own faster. In fact, a lot of a natural evolution and progression we're seeing is people using their personal Gmail accounts to figure things out and then slowly bringing in what they've learned into the office. One of the things that really smart organizations are doing that some that are perhaps lagging behind aren't is actively embracing that external experimentation to internal new process adoption funnel. organizations and enterprises are constrained on how freewheeling and experimental they can be.

Starting point is 00:14:15 However, I would say that in general, they overestimate risk, and they undervalue finding ways for people to experiment even if they can't do it with corporate data. The last thing that I will say is that I think that one other outcome of this progression of technology diffusion that Andre is noticing is that we are likely to see a radical increase in bottoms-up entrepreneurship. The amount of activities that seemed inaccessible to people before or capital constraining because they would have had to hire someone to do it, and which stood in the way of them just doing the damn thing that they've wanted to do, has decreased radically.

Starting point is 00:14:45 And that means the number of people who are going to just throw off the shackles of their normal enterprise and try to do the thing that they've always dreamed of is going to increase. I think we're going to see an absolute Cambrian explosion of small business entrepreneurs, solopreneurs, people building stuff because that becomes easier. And that's one of the impacts that I'm most excited to see play out. Anyways, big thanks to Andre for another thought-provoking peace, and thanks to you guys, as always, for listening or watching. Until next time, peace.

The AI Daily Brief: Artificial Intelligence News and Analysis - Andrej Karpathy on How AI Empowers

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.