blank_crypto

🚨 DFINITY Foundation has begun sending out "cloud engine" coupons. We're going slowly, but will go faster in coming weeks, as we process a large waitlist. The timing of the massive Microsoft outage is coincidental, but current happenings are part of the why. 🧵

ICP "cloud engines" launch trajectory: DFINITY will soon begin disbursing coupons with enough cycles to create and power an engine for 1 month+, unlocking in sequence: 1. Agentic infrastructure ⚡️ Quickly build secure, sovereign onchain production apps using Claude Code, Codex, Caffeine, ... 2. Open SaaS beta release ⚡️ Install AIware apps that can run any enterprise, as easily as installing phone apps... 3. Unlimited growth ⚡️ A new end-to-end, mainstream onchain cloud tech stack will come into being. *** What this means *** ICP cloud engines are a new development on the Internet Computer designed to allow it to play the role of a revolutionary new cloud tech stack for the AI era. Currently, the Internet Computer scales by combining internal subnets, which are akin to individual, high performance blockchains hosting secure serverless software called canisters (something like smart contracts, but much more powerful, which can be used to create huge onchain apps, which serve their web screens directly to users) Cloud engines allow enterprises to create their own subnets that belong exclusively to them, which they control and configure — while their canister software can still seamlessly interoperate with software on the main network, and engines run by others. An enterprise creates a cloud engine by choosing nodes from a marketplace, which the ICP protocol combines into a revolutionary serverless cloud platform for them, where the computation and data of their hosted apps and services is tamperproof (immune to infrastructure hacks) and always on (incredibly resilient), AI agents can build and update apps far more effectively than on traditional tech, and next generation AIware software can be created. Today, enterprises demand to be able to control where their sotware and data lives, and cloud engines make that possible with no loss of flexibility. Cloud engines can even be re-configured with new nodes, without interrupting the hosted apps and services — the ICP protocol just handles it, granting complete sovereignty and freedom from vendor lock-in. In advance of general release, cloud engines are already being used by governments and NGOs seeking to create sovereign cloud tech stacks and apps. On general release, they will be used in far more mainstream ways too. Here are the details of what's coming. *** Agentic Infrastructure *** In the future, and really already today, nearly all of our apps and services will be created by AI agents rather than humans. Further, next generation software is "AIware," which has AI deeply integrated on the inside, which allows it to provide superpowers relative to traditional software. Agents have different needs when they building things. They can write software orders of magnitude faster than humans, and we will demand they update our production apps and services at the speed of conversation. Functionality will be fluid and its shape will change to match what we want. On traditional tech stacks, updating apps and services is a necessarily laborious and risky process for three reasons: 1) a security flaw might be introduced, 2) a systems configuration change might break them, and 3) the way that data is migrated might cause some to be lost, often in subtle ways that aren't detected at the time of the upgrade. This means that AI agents can't just be let of the leash to build and update at the speed they can. ICP cloud engines solve this. Hosted software is tamperproof, so agents can update without needing security teams in the loop. Hosted software is serverless, and the data lives in the logic, so apps don't. need things like operating systems and databases configured to run, so agents can update apps without having sysadmin teams in the loop either. And, if Motoko 2.0, the world's first languge targeting agents is used as the backend language, its framework is able to detect when an app update by AI might cause data loss, and simply require the agent writing the code to try again, reducing potentially catastrophic data loss to a harmless retry. Crucially, apps built with Motoko can also put AI agents on the inside, so that subject to permissions, they can see the data graph in the persistent memory used by serverless software on ICP, and access its functions. This enables you to instruct agents to do things in the apps for you, such as analyzing data, or performing actions, with great flexibility and power. ICP cloud engines are agentic infrastructure that change the game — and it will be possible to use them as such from the moment they are released. There is already a government using internal production software built on a cloud engine using Caffeine. But cloud engines will meet a wide range of mainstream needs, from solorpeneurs to startups, SMEs to corporations, governments and NGOs. *** Open SaaS *** The obvious question, is if AIware is so much better, shouldn't all enterprises be running on it? We believe they should, and that his must be enabled. This has been a labor of love for me. A new venture associated with me will release the Open SaaS suite for use on cloud engines once they have been running a short while, which suite, as you've probably guessed from the name, is open source and free. Open SaaS is a suite of every important SaaS (Software-as-a-Service) app used to run enterprises around the world today, reimagined as onchain AIware that runs on ICP cloud engines. Because of the unique nature of the software that ICP hosts, installing and maintaining these sophisticated enterprise apps on a cloud engine is as easy as installing and managing a phone app. A small selection of the apps: Open Team, Open HR, Open Accounts, Open CRM, Open Projects, Open Email, Open Messaging, Open Signing, Open Knowledge, Op... etc. There are many more, and there will be surprises. The user experiences are beautiful, and easily compete with the big money SaaS screens many enterprises are used to. Moreover, the are linked to what I called a "software factory" driven by AI, and the more enterprises that use them and provide feedback, the better they will get, in realtime. Today's expensive SaaS silos make it hard to integrate and combine enterprise data, and to unleash the full power of AI. They are the opposite to Open SaaS. In the future, enterprise users will spend far more time in AI chat than on SaaS screens. That's the future that Open SaaS enables and unlocks. Enterprise users can even say to their Open SaaS suite, "make me more profitable," and it will have a go. The big surprise coming for the world: once you run your enterprise on AIware, and AI can access all your data in realtime, and is integrated with your software's actual functions — it can become as good at running your enterprise as writing software for you. It's revolutionary, and ICP is pioneering this. Not a narrative, but real decentralized network tech. Something I can share: DFINITY is in the process of migrating to Open SaaS from a vast suite of expensive SaaS silos, and is already reaping big benefits as we refocus ourselves for the era of AI. **** What does this mean for the project? I believe ICP cloud engines are the vector that will transform the the Internet Computer into a mainstream agentic infrastructure tech platform for enterprise— which direction platforms like Caffeine have already helped the network travel, by enabling many tens of thousands of users and enterprises to create onchain apps on the shared network through online conversation. Now, the next and most important phase is approaching. Our go-to-market will proceed quietly as we rollout coupons and work to fettle the new network functionality, but I think that in hindsight, the launch of cloud engines will be seen as the most pivotal moment in Internet Computer history since network launch. With this, ICP shall show the industry how networks can provide real agentic infrastructure. Bear with us as we tie up loose ends. We never stop working, and you can be sure that we're trying to get this out as fast as possible 💪⚡️💥

The opencloud.org portal is designed to be autonomous and run as an integral part of the Internet Computer network. It will thus be updated by the NNS, the network's inbuilt decentralized governance system = sovereign frontier cloud from an open network.

Traditional tech security is cooked with AI. Limitless combinations of clouds, servers and platform software configs. Zillions of packages auto-downloaded by e.g. Node.js. A vast surface where vulnerabilities hide, which can now be found Hackproof stacks soon to be *necessary*

Internet Computer "cloud engines" will run over AWS data centers i.e. ICP will provide a way to create full stack apps on AWS that keep running when individual DCs have issues. AWS has been very helpful, and provided technical hooks needed by ICP node software. Coming soon.

DeepSeek has showed us LLM training can be done with 5% of the hardware. That got the attention of hardware manufacturers like NVIDIA, and big hardware buyers/investors like OpenAI. But: there may be a NEW source of demand for hardware for AI, albeit, for a different kind of hardware... (Many are questioning whether DeepSeek really has massively reduced the hardware needed for training and inference, sometimes suspecting Chinese state trickery, but looking at the technical material they've released so far, at least, everything looks real – and denial is a normal reaction to disruptive advances in tech, as we found out with ICP) The insight regarding hardware trends I will share comes from DFINITY's work training LLMs for use with the "self-writing internet" paradigm, which, in our case, requires them to be great at writing code in the Motoko programming language, and understand the Internet Computer environment, among other things. (For those unfamiliar with the background, the DFINITY Foundation is developing public infrastructure where AI builds and *updates* sophisticated sovereign web apps and internet services working completely solo, purely on receipt of natural language instructions.) In our work, training involves fine tuning different foundation LLMs (i.e. taking foundation LLM models like DeepSeek, which have already been very broadly trained, and then finessing them with regards to specific knowledge), and benchmarking the results. To perform fine tuning, we create special training data, such as new Motoko coding examples, which demonstrate how to correctly solve a wide range of coding challenges. This material has benefits beyond fine tuning of course. Once it has been proven in fine tuning, then we can a) upload it to the internet so that it can be found by those creating/pretraining foundation models, magnifying its effect and reducing fine tuning needs, and b) stored into something called RAG, which is kind of infrastructure that acts as memory for AI, and can provide it with concrete context during inference tasks. Naturally, developing this material is time-consuming and involves significant human resource. Even though we are rapidly scaling our AI team, who are brilliant people, we want to go faster! Therefore, at some point, we started experimenting with using AI to create additional synthetic training data. Now, this might sound like a dangerous recursive approach that's equivalent to "getting high on your own supply," but in fact, it has already proven somewhat effective. Meanwhile, we think that a new development in the science of LLMs called COT, or "Chain of Thought" reasoning, may supercharge this approach. With COT, rather than use more training data and hardware to make LLMs smarter, we have them spend compute resources reasoning about inference tasks they have been given by breaking them down and evaluating different lines of thought to produce a result, than just blurting out what sounds best based on their training. The more time spent "thinking," the better the result. DeepSeek has show how good COT can be, and here's the rub: in principle, by dedicating more compute time resource to reasoning, we can create high-qualify synthetic training data that is nearly free from hallucinations (which would indeed be recursively bad if they were used for training), where the quality and "originality" effectively derives from the money spent on compute. Such synthetic training data, once produced, can then be reused in the pretraining and fine tuning of any models we wish to make smarter, and can be used over and over again without limit because it is just another form of digital content. As per scaling laws, generally the more data available for training, the better the result. So this is a ratchet of sorts where compute is the lever. So this is what I think: 1) Using COT to create high-quality synthetic training data consumes a lot of hardware time, but nonetheless costs far less than using humans to create the same kind of data, and; 2) COT can produce new training data much faster than humans, and; 3) We will now need vastly less hardware for pretraining, as shown by DeepSeek; 4) So we can instead invest more in using COT to produce powerful synthetic training data, which will allow pretraining to create smarter models, and; 5) For training purposes, we are going to want far more hardware optimized for *inference* that creates this data, and far less that is optimized for the act of pretraining itself. Plainly put, while many have been investing heavily in hardware that is optimal for pretraining, such as H100s, the world may need less of that, and more hardware that is specialized for inference to pump a synthetic training data ratchet. Being more specific: arguably, therefore, the world might need more hardware from companies like Groq and Cerberas, and less from companies like NVIDIA (even though I'm sure its order books are full well into the future for now). Moreover, AI companies who have not already fully committed their hardware budgets, and are able to pivot to purchasing inference hardware, may be able to obtain significant advantage. Maybe some big players have jumped too soon. If so, this will really upset the apple cart. It's too early to see how this will play out, of course, and none of what I have said is certain. However, AI is a highly dynamic space, and startups should see opportunity remains around every corner, despite the huge amounts of capital already deployed. DeepSeek, and changing calculuses, should also put the frighteners on Big Tech in the Western world. @pmarca is right that DeepSeek is AI's "Sputnik moment." Our leadership in AI, and assumptions about why we can maintain it, have just been severely shaken – we'll need to raise our game.

By popular demand, here's a detailed post about caffeine.ai and the forthcoming "self-writing internet" (SWI) paradigm we in the ICP community are working to realize 🌐🧠 If you prefer pics, checkout my deck at deck.icp.ai. Otherwise here goes... Caffeine is meant for everybody. Using natural language, anybody will be able to create sovereign websites, web apps and internet services. It might be an individual wanting to create something like a personal branding website, wedding planning app for their family, or portal for a sailing club. It might be an entrepreneur lauching an e-commerce website, or a sharing economy app with advanced Web3 rails. It might be a company or gov. dept. that needs a corporate portal, or some specific CRM (Customer Relationship Managment), ERP (Enterprise Resource Processing) or ARP (Accounts Receivable Processing) functionality. The SWI paradigm is broad, and it involves anyone and everyone being able to create and own anything online from scratch, just by talking to AI. Over time, as the paradigm becomes more and more powerful, this will increasingly empower, democratize and decentralize tech worldwide. Today, Large Language Models (a type of AI) have already become very good at writing code, and their abilities are continuing to improve rapidly. The Caffeine platform leverages this ability in combination with new specialized SWI infrastructure and technology that is being deployed to the Internet Computer network. This new tech infra has been especially designed to empower and enable AI to build effectively and safely solo on receipt of natural language instructions, without receiving any technical assistance from humans. Caffeine has also been carefully designed to allow technical people, such as engineers and user experience designers, to build alongside AI, with purposes such as accelerating time-to-market, dramatically reducing costs while increasing quality, and using the AI as a tutor to learn new skills. It's important to know that AI building apps solo is only one part of the SWI paradigm. It also involves apps and services being created on the decentralized internet, rather than inside some company's proprietary centralized platform, and drawing on this difference to provide 4 very important, defining characteristics: 1. Sovereignty The SWI paradigm produces apps and services that become accessible via URLs (the addresses we enter into web browsers) that are *sovereign.* This property results from them being entirely hosted (running) on a decentralized, public network, in exactly the same way a bitcoin token lives on the Bitcoin blockchain. They are constructed from secure network-resident code/next-generation smart contracts, without any centralization or traditional backend IT needing to be involved, such as cloud computing services. This means that apps are not dependent on intermediaries that might censor them, or which might inflict harmful actions like massively hiking prices, which services typically seek to make them captive customers by locking them into their platforms. In addition, the underlying software and data of apps created by SWI belongs exclusively to their owners (which might be humans, an organization, DAO or yet another kind of entity), further preventing intermediaries gaining leverage. In production, these sovereign apps are controlled by a wallet (or assigned to a DAO, say), rather than a cloud account operated by a company. 2. Tamperproof security Those building using traditional IT often euphemistically say their web apps and services are "secure," because they have carefully configured their cloud accounts to prevent exploits, and have purchased cybersecurity measures to protect hosted platform software like databases and their traditional custom code. However, in reality, such traditional IT infrastructure can never be truly secure, since the cybersecurity protections involved are highly fallible. In complete contrast, the websites, web apps and internet services created on SWI are "tamperproof," which means they are fundamentally secure, and don't need traditional cybersecurity measures for extra protection. This makes it much easier for AI to build solo, and also frees app owners from having to trust that centralized traditional IT service providers and SaaS platforms will successfully keep their data and systems secure using cybersecurity systems, which they cannot guarantee even if they wish to, because traditional IT is fundamentally insecure, and can only be patched by cybersecurity measures. 3. Updatable in production The SWI paradigm requires that the owners of the apps and services produced can update them almost in real-time, simply by issuing additional natural language instructions to an AI, essentially evolving them at chat speed (in practice, with owners issuing instructions, and then refreshing the URL where their app is located). If you are non-technical, I need to tell you that this requirement is far more challenging than it might sound. Firstly, the process of upgrading/updating apps running in production must be sufficiently swift to occur at chat speed, which requires that the data involved can be transformed into new formats quickly. Secondly – and this is the really tough one – it is required that when the AI modifies the underlying software logic, and performs any required transformations/migrations of the app's data, that no data is accidentally lost, for example because the AI has made a mistake, which LLMs will continue to do because of the way they work, even as their overall abilities continue to improve. 4. Trustless Web3 SWI apps can incorporate many different types of functionalities, including AI functionalities. But there is a form of functionality that cannot be safely implemented on traditional IT stacks: Web3 functionality. SWI apps can securely hold and process tokens on multiple blockchains, trustlessly and seamlessly incorporating them into their own functionality, without requiring users use cumbersome traditional crypto wallets, which undermine user experience, and are themselves are vulnerable to hacks. The Internet Computer, in combination with the traditional blockchains it is directly interoperable with, forms a World Computer. The Internet Computer already hosts many onchain services that process tokens from multiple traditional chains, without the involvement of bridges and other systems that are often centralized in practice (note that in Web3, oftentimes people refer to apps built in a centralized way using traditional IT, which run on servers and compute platforms like Amazon Web Services, as "onchain," simply because they have an associated token, but in reality of course they are not, and therefore, cannot provide these guarantees). A great example of what's possible with the fully network-resident paradigm is , OpenChat, an open internet service that runs on the Internet Computer, which enables its users to instantly send bitcoins, ETH and SOL inside chat messages (the Internet Computer's "chain key" technology enables it to play the role of a kind of Layer 2 for blockchain assets). SWI will be able to build apps and services with similar functionality. Where required, those apps can also maintain their own treasuries using stablecoins (e.g. tokenized dollars) and other crypto innovations. DFINITY is performing a wide range of technical work related to SWI. For example, we have an AI team that works with LLMs to make them better at coding SWI apps using specialized training and RAG infrastructures, and we are working to make efficient, seamless ICP network scaling even better, in anticipation of increased demands on the network created by SWI, and many other things, but in this post I'll talk more about our Motoko work. Technical note (skip don't balk!): The Internet Computer hosts an entire serverless cloud, in which canister smart contracts are hosted and run, within a "virtual execution environment," which leverages the WebAssembly/Wasm virtual machine. (The virtual execution environment is hosted with the network's stateful decentralized network protocol, which is called Internet Computer Protocol, or ICP, which the network's sovereign node machines process). This means both human developers, and AI, can write code in any programming language that compiles to Wasm byte code. Several languages are now popular in the ecosystem. However, back in 2018, 3 years before the network launched, DFINITY decided to develop a completely new programming language for ICP platforms, called Motoko. Aside from applying advanced emerging concepts in programming language design, we were pursuing a number of objectives with Motoko: 1) To create a new, broadly accessible, simple-to-learn language, that was nonetheless powerful enough to build complex systems (addressing shortcomings of simple languages like JavaScript). 2) To create a language that leveraged special aspects of the Wasm virtual machine standard. 3) To create a language that was specifically designed for the Internet Computer's virtual execution environment, and some of the incredible advancements in the field of compute that it provides (in technical speak, Motoko is DSL/domain specific language). Motoko is far more sophisticated and powerful than something like Solidity or Move, and developing a new language from scratch is not a simple undertaking, but we set to the task. We were no doubt helped by the team being initially led by Andreas Rossberg, a co-inventor of WebAssembly, and leader within the Wasm community, who joined the DFINITY team in early 2017. He helped form a brilliant languages team, which continues to carry the torch today. At some point, we realized that when AI builds solo, it might benefit from a language like Motoko, and thus, critical aspects of the language's design are being developed to empower human and AI developers alike. Prima facie, without any special training, LLMs coding using Motoko already make far less mistakes than using Rust, an increasingly popular modern systems-programming language – despite their being far more training material available for Rust, in the form of online documentation, and open source code repositories. But by developing special materials for training LLMs, and new technologies for Motoko, we are taking this much further. One of the key advancements provided by ICP, is that software runs within a compute environment that providing something called "orthogonal persistence" (but don't balk, this is easy, keep reading!!) This essentially means that software logic runs within persistent memory pages, and it's no longer necessary for developers to use database servers and file systems to persist data. In a sense, logic and data have become one, because developers can just define logical abstractions, and the data then automatically persists within them. In summary, orthogonal persistence: 1) Massively simplifies software design, making it more abstract, which greatly helps AI. 2) Improves performance, since application data no longer has to be copied in and out of databases and files (this is part of the magic that allows, say, a social network to be built on the Internet Computer, while there is still not a single traditional blockchain that cfan store a simple photo taken with a phone). Motoko development is leaning into orthogonal persistence, and I expect within 10 days, will release "enhanced orthogonal persistence," or "EOP," which takes things further still, especially with respect to AI empowerment. With EOP, when canister smart contracts written in Motoko are upgraded, where the structure of data does not need to be transformed, it can be left completely alone during upgrades. This is another simplification, which will greatly help human and AI developers alike. But there's more... When the SWI paradigm creates complex web applications and services online, inevitably changes to the structure of data in memory may occur. For example, image an e-commerce website, with a Product object. The owner of the site might decide they wish to add comments functionality to listed products, which would require a Comments field being added to Product. At this stage, during the upgrade, the old Products objects must be transformed into new Products objects, and with EOP, some migration logic must thus be created. EOP ensures that transformations can occur quickly, which is crucial for allowing SWI apps and services to be upgraded at chat speed, but it also addresses an even more important need. Due to their design, LLMs will always occasionally hallucinate, and even though they are smart, they will make mistakes (and considering they are coding hundreds of times faster than humans, this is no surprise). What then if an AI building solo makes a mistake in the migration logic, causing app or service data to be lost? For example, what if the AI defined the new Product format, adding the required comments feature, but in the process caused previously entered product data to be lost? (Of course, Caffeine will provide snapshotting and restore functionality, but this is last resort, not something a non-technical developer simply talking to build will want to use.) The answer is that in EOP, when an upgrade is installed, the Motoko language platform checks the migration logic against the currently installed version to check that no data will be lost – and simply fails if it would be, forcing the AI to try again. This is called "loss safe" programming, and it was invented at DFINITY. I understand this is a bit of a technical story, but hopefully you can see that the "updatable" requirement of SWI is challenging to implement in practice, and thus that EOP delivers another groundbreaking ICP advance that will help AI and human builders alike. None of this comes easy. Many aspects of EOP were envisaged back in the early days, but it has taken many years to realize the technology. What is clear, is that AI must be empowered to build solo, because that is the future: AI is not human, and in the SWI paradigm, AI is required to build and update apps and services extremely rapidly, at the speed of chat. To make all this possible, the bottom line is that AI requires its own environment to build in. ICP already provides a tamperproof compute environment, so that a mistake by AI cannot lead to a hack and data exfiltration, or the installation of ransomware say – it is truly secure. ICP already provides an unstoppable compute environment, relieving AI of the need to create things like failover and backup and restore systems, massively further simplifying software logic, enabling the AI to focus on what, not how, and further reducing the risk a mistake might cause data to be lost. Now, Motoko's EOP will further accelerates upgrades on ICP, allowing them to run at chat speed, and most importantly, creates guardrails that prevent a simple mistake by AI causing data to be lost. Of course, this is not all, and over time, I will have more to share with you about our SWI work. What I hope I have conveyed, is that SWI is a new and emerging technical field, which is one that the Internet Computer, and the ICP community, is uniquely positioned to capitalize upon In the future, human developers will build alongside AI developers (to reduce friction, we are also developing an in-browser development environment, called icp.ninja). But also in the future, all 8 billion humans on earth without technical skills will also gain the ability to develop and own tamperproof and unstoppable sovereign websites, web apps and internet services, just by speaking in their local natural language. When we release Caffeine, that will be only a beginning of a long and incredible journey into the future. The SWI paradigm has no limits. As AI gets smarter, and as Caffeine, and the underlying SWI technology and infrastructure on the Internet Computer improves, it will be possible to create increasingly advanced apps and services. Individuals, entrepreneurs, businesses, and governments, will increasingly choose to leverage SWI to create sovereign custom apps and services that are completely under their control, that they can update, customize and improve just by talking at any time of their choosing, and which slash their time to market and costs by orders of magnitude compared to today's tech – and the total addressable market will become unimaginably vast. We want the SWI to explode on the Internet Computer, and are working towards that goal. More news soon 💪

This is pretty (very) cool.

AI x ICP x internet x everyone (self-writing internet)

The AI x internet revolution coming on #ICP: *Self-writing* sovereign web applications and internet services that owners can update in real-time by talking 💥 The Internet Computer has been designed for this and work has been ongoing for YEARS. Today I want to explore this topic in a bit of depth for the first time. Please note — this area of work is not to be confused with the AI capabilities of ICP networks (for newcomers: the Internet Computer is unequivocally the ONLY public network in the world that can host and run AI in the mode of smart contracts, which makes it network-resident and decentralized, and secure and unstoppable, and for example, I recently demonstrated onchain neural networks performing facial recognition, and pending enhancements to the ICP protocol will allow LLMs to run as smart contracts too). Today I'm talking about a very different challenge that ICP shall help the world meet, where *running* web applications and internet services are created and updated simply by talking. Users will create these for any purpose – to create something like a secure personal notes manager, or a personal website, or something for others, like social media, games, web3, or enterprise infrastructure, which involve online communities large and small – simply by issuing instructions in natural language. Please also note, this is a far bigger and different challenge to using AI to write and review software, which is already happening at scale. This is something well beyond, for reasons I will explain... To begin, let's step back and look at some general trends, to understand where the internet is going with AI: Many of you reading this, will already be using ChatGPT to explore ideas, get information, and analyze, refine and create content, and soon to search the internet. ChatGPT is an example of a large LLM (i.e. chat bot) with a huge number of parameters that has been trained on a colossal amount of data. If you're a software engineer, you'll already have been using LLMs to help write and review code, although maybe you prefer Claude. These models triggered the recent wave of hype around AI. But in fact, they are manifestations of a deeper trend that AI has been driving for some time now. The deeper trend is this: we interact with AI, knowingly or not, and it gives us something we want. Some of the first mass market manifestations were services like TikTok and Instagram Reels. These services are not traditional social media services at all. Under the skin, they are driven by powerful AI engines, without which they wouldn't work. When you use them, you are really interacting with AI. The AI inside these services classifies the social media content they can serve, so that it knows what's inside videos and other kinds of posts. Then, as it feeds you content to consume, telemetry is collected, enabling the AI to track how you interact with the content – for example, most simply, it can track how much of a video you watch, to determine what kinds of content you like. As the AI gets better at understanding what you find compelling (which can extend to the sequences of videos it shows you, not just individual videos), you receive a better and better experience, which is why these services are so incredibly addictive. This new media model is just part of inexorable journey we are being taken on by AI technology, where it gives us what we want – which will be far reaching. Seemingly impossible things are now happening. For example, researchers recently trained AI on millions of hours of footage of people playing Doom and Minecraft. This has enabled the AI to simulate those games in realtime for users. You can play the games, but there's no game server or game client, just an AI streaming video to you in response to traditional inputs such as left, right, run, jump and shoot (to be clear, there's no game server, or game client, or any other infrastructure involved from the original games). This hints at what's coming: in the future, AI will create virtual reality experiences for us, and watch how we interact with them, tailoring them in real time to make them more enoyable and compelling. Sci-fi as this might sound, really will just a continuation of the trend exemplifed by TikTok and Instagram. But enough of this. What else will AI give us on demand where ICP specifically plays a role? We in the ICP community are heavily focused on reinventing the platform on which we build, specifically providing a better way to create web applications and internet services (including those with web3 and AI functionality inside), which also happen to be sovereign. In the future, we will just talk to AI, to create and update the *running* web applications and internet services we want too. It's obvious why. For example, what if a businessperson wishes to build a custom personal website, to promote their brand, with blogging functionality, a section where they can embed media pulled from places like YouTube, a library page hosting documents they have created, and a page linking to their social media profiles – and have all that functionality look a certain way. In the future, will they still be expected to hire designers and developers to build such a thing, or otherwise be stuck fiddling with Wix, or just sticking with their vanilla LinkedIn profile page? Of course not, they'll just talk to AI. What about an avid gamer, who wishes to create their own custom online game to share with friends? Will it still be difficult to express creativity without special technical skills and a lot of time? Nope. And how about a business, NGO or government that just needs some custom CRM functionality? Will they still have to sign-up to big-ticket SaaS services like Salesforce, and hire consultant to customize them? Today, creating things on the internet is complex, time-consuming and costly – and this stops us building what we want. The world is waiting to break free from this. In the future, we will just talk to AI to create and update, which will result in an almost limitless number of new custom web applications and services being created for every purpose imaginable. Here's how it will work: 1) You'll describe to AI the custom app or service you want 2) It will return a URL to your web browser, and there it will be, ready to use! 3) You and others will use the app, causing it to accumulate content and data. 4) You'll describe improvements, extensions and fixes. 5) Then you'll just refresh the web page to see them. 6) Loop to 3, and continue iterating to realize value. Over time, this new paradigm will massively change the way tech works. Imagine what this might mean for budding entrepreneurs around the world, who lack the technical skills or the funding to hire them, but who have ideas for ventures in social media, games, the sharing economy, AI-based services, web3 services, communication suites, and _ (fill in what you want). This will democratize access to the tech economy, and help enable vast additional worldwide talent to participate and become industrious and successful. This is an aim that has been part of the ICP project since inception. Furthermore, this functionality will be for all humanity, not just entrepreneurs. Imagine a simple group of high school students that want to organize the information they collect from a biology field trip. It works for them too! Imagine some business department that needs custom online functionality, but can't convince the CIO and CFO to allow them to sign up for something like Salesforce and hire some consultants to customize it. (Which will anyway take a very long time, and cost a huge amount, even if they agree). Solved. Now think about the current situation in the developing world. They having growing needs for custom online functionaltiy, but can't afford Big Tech SaaS services, and meanwhile don't have the skills to build it out themselves – and if they did, then they would also need the cybersecurity expertise to make what they built safe, which they lack. For those economies, this will be trasformative. The new paradigm will solve for all these needs, and in fact go one better. People who create custom applications and internet services will OWN the software that makes them, even though they don't write it themselves, and OWN the data inside them – which contrasts to popular SaaS services businesses use, which holds customer data hostage, and even consumer services like Google Photos, which make it impossible to get media out. These custom applications and services will be truly sovereign, and the owners won't be captive customers – which has long been an aim of the Internet Computer network. Counterintuitively, this new paradigm will also be great for software engineers – it will result in millions of new custom applications and services being created, and inevitably, there will be places where human assistance is useful to solve specific issues, and help making prompts go further. The orders of magnitude increase in online infrastructure will create huge numbers of software engineering jobs, all around the world. If you have stayed with me this far, hopefully by now you can see that this new paradigm is both inevitable, and represents one of the biggest revolutions ever in tech. So... the next question is, how exactly will ICP finally uncork the paradigm for the world. To understand this next part, we first have to look at the limitations of traditional IT when applied to this paradigm. Creating and updating *running* online applications is much more complicated that getting an LLM to write some software code. For example, using a typical traditional IT framework to build, this is what might be required: 1. Get an AWS account, and add a credit card 2. Get some "compute instances" (effectively servers) 3. Install some cybersecurity to make it safe 4. Install a database server, a web server, a... 5. Orchestrate it all with e.g. Kubernetes 6. Patch all your software for security 7. Design failover, backup and restore 8. Create tables in the database 9... 10. Install the code involved This is quite a list of tasks, some of which are very involved, so the AI will have to do much more than just write code. Let's assume the AI is given hooks, and has the knowledge, to perform all these steps itself. Will this solve the paradigm? Unfortunately there will still be issues... At a basic level, the paradigm should provide users with a real-time creation experience, and even installing a database server or patch can take some time. Of course, this kind of thing can be mitigated by using preinstalled images, but problems run deeper... The various steps and requirements involved in traditional IT can go wrong in multitudinous ways. The AI's build sequence will break, like it does for humans, and it will have to make judgements about how to address them within an *unconstrained problem space,* which will subtly impact things like security – which is a serious problem, because traditional IT is insecure by default, and even small mistakes can result in disaster. Traditional IT is a crazy complex Rube Goldberg machine, and providing AI with sole responsiblity in that kind of unconstrained problem space is potentially very dangerous, because it hallucinates, and can pickup bad memes from its training data. Everything the AI does would have to be manually reviewed and, if the application or service is important, audited, by technically competent people – and of course – the whole point of this paradigm is that it doesn't require creators to have technical skills, and to make creation a real-time iterative act. And there are other showstoppers using traditional IT. In the new paradigm, the users/creators will want to update their running web applications and internet services in real-time, simply by telling the AI about the improvements, extensions and fixes they want. The systems created will need to undergo substantial upgrades every few minutes!! Traditional IT has not been designed to make this possible. Anyone involved will know that upgrades are a big deal, and with production systems, typically occur on long time intervals. That's because changes often have to be made across multiple components in a synchronized way (e.g. updating database tables, changing web server configurations...) and it's a fiddle. Moreover, when you change the design of web applications and services running on traditional IT, the upgrade process often involves refactoring/migrating data, which process is slow, computationally expensive and error prone – again, preventing traditional IT realizing the crucial real-time nature of the paradigm, which will involve running web applications and internet services being updated almost at the speed of chat. I could go on, but it should become clear traditional IT is not really suitable for the paradigm. Because of the difficulties, we will see services like Vercel, and probably eventually services like Google, providing AI that can create apps within custom infrastructure platforms they craft, which ameliorate some of these problems. But their platforms will remain less than ideal, and moreover, the software created by the AI will also be locked to their special platforms, and they will probably also hold the data involved hostage somehow, creating customer lock-in, and the applications and services involved won't be sovereign. (Nonetheless, we predict that certain web3 projects that tend to imitate ICP, in desperation will use something like Vercel to create dumbed-down versions of this paradigm, then mis-sell their services to the token-buying public as being "onchain," but beyond successfully selling more tokens for them, their schemes will ultimately fail to be competitive with mainstream users/creators worldwide.) So what is really needed? The good news is that DFINITY has been working on a solution to this paradigm for years already. Let me explain... From the beginning, going back years, and now more than a 1000 person-years of R&D effort, our work has been ruthlessly focused on reinventing compute broadly using a decentralized network leveraging advanced cryptography, protocol math and computer science. Our work is completely unique in the web3 field. A key feature of ICP is that you can build a web application entirely from secure and unstoppable netwok-resident software, which a vastly more powerful evolution of smart contracts. When you build on the network, you don't need Big Tech and traditional IT. AI can write this code, and upload it to the an ICP network like the Internet Computer to create an web application or other internet service. To be clear, in the radical compute environment ICP creates, AI can create and update, just by uploading code... There is no need to configure a cloud account, a database, a web server, cybersecurity. Moreover, the code is automatically secure and immune to cyber attack. So there's no need to worry that a hallucination will leave a door open for hackers. Moreover, there's no need for the AI to design and configure complex failover systems, because the code is unstoppable and always runs. So those roadblocks are out of the way. But the real power derives from a seminal computer science advance that ICP provides called "orthogonal persistence" (keep reading, I won't go too technical on you!) On ICP, units of software run inside persistent pages of memory, which is basically to say, that data "sticks" to the software logic that programmers create when they write in a software language, which saves them from the complexity and inefficiency of copying data in and out of databases and files, also removing the need for those things. Everything is just network-resident highly-abstracted software. This makes it possible for engineers, and soon AI, to describe functionality in a much more simpler form, without dependencies, which is ideal for this paradigm. I described a vision for "orthogonal persistence" back in 2018, but only now, 6 years later, will it be fully realized through Motoko, a domain specific language that directly hooks in to the workings of the ICP platform. Get ready for what we are calling EOP, or "ehnanced orthogonal persistence," which will finally realize what we have been working towards. (This is gated by pending upgrades like the 64-bit change also need to run LLMs on the network, which are about to happen.) Above I talked about the importance of instant and safe upgrades to the new AI paradigm. Well.. EOP makes it possible to "morph" software between upgrades. Developers (human and AI) will write new versions of software, which realize the changes required. Then separately they will describe code that transforms that data from the old version. (For example, if a Google Photos-style application has been created, then an upgrade might add location data and comments to photos, causing the strucutre of the "photo" data type to change.) During upgrades, in the new paradigm, EOP does the following: 1) It adds type safety to the upgrade, ensuring that if the AI has made a mistake that might cause data loss, by hallucination or otherwise, the upgrade will fail, which greatly reduces the risks that would be ever-present in traditional IT architectures. 2) As software is morphed through upgrades, it allows data transformation to occur in a highly efficient manner, such that the paradigm can deliver real-time upgrades at the speed of talk. Which is what is needed. Other benefits from the years focused on reinventing compute are important too. For example, because code and state exist as one within this the environment, ICP can easily almost instantly snapshot applications and services, enabling users to rollback to precisely where they were before, if they don't like how their data was transformed (using EOP, this can often also be done just by "upgrading" to earlier versions of the software). I could go on, but I'll summarize. This new paradigm, which will ultimately profoundly change tech, will be unlocked by combining ever-improving AI, with ICP technology. What's incredible for the ICP community, is that the paradigm provides utility to massive worldwide market, and we will not be constrained by web3 noise. People will use it because it gives them what they want. I can tell you that behind the scenes we are expending great effort on AI itself, and also on the framework that will let it build on ICP, and on making sure that the Internet Computer can scale to handle this – which work you might already be aware of. If you think the incredible recent growth of compute on the Internet Computer is impressive, buckle up, because this paradigm may mean we haven't seen anything yet. As always, we choose to believe in this over narratives: Pure. Utility. From. Advanced. Alien. Tech. Will. Win. In. The. End. Today we are closed than ever. O, btw did I mentioned that this next generation of web applications and internet services that people create using AI will be internet native, and sovereign. They will run on a network hosted by real decentralized hardware (the Internet Computer is one of the few web3 networks that does not really run on Big Tech), leverage trustless multi-chain functionality, and embed real onchain AI. This is going to really, really cool...