Open vs Closed Models

write a long form blog post in my voice about : https://www.interconnects.ai/p/open-and-closed-models-are-on-different - make it relate to my present work with https://github.com/kliewerdaniel/objective05.git

Here's a draft in a style that feels aligned with the way you've been writing lately — tying Nathan Lambert's argument directly into Objective05 and your broader obsession with local-first AI systems. :::writing{variant="document" id="58271"} # Open Models and Closed Models Are Solving Different Problems I recently read Nathan Lambert's essay, "Open and Closed Models Are on Different Exponentials," and it articulated something I have been slowly discovering while building Objective05. The mistake many people make is assuming there is a single race. They imagine OpenAI, Anthropic, Google, Meta, DeepSeek, Qwen, and everyone else running toward the same finish line. The assumption is that eventually somebody wins, everybody else loses, and AI becomes another platform monopoly. I do not think that is what is happening. Nathan argues that closed models and open models are operating on different economic curves. The frontier labs are optimizing for maximum intelligence. Open ecosystems are optimizing for diffusion. The frontier labs are selling leverage. Open ecosystems are selling ubiquity. The frontier labs want to build the smartest possible employee. Open ecosystems want intelligence embedded everywhere. citeturn0search0turn0search4 The more I work on Objective05, the more I think he is right. Objective05 is not an attempt to compete with frontier models. That would be absurd. A handful of companies are spending tens of billions of dollars training increasingly capable systems. No independent developer, startup, university, or hobbyist community is realistically going to outspend them. But that observation immediately leads to a second question: What if intelligence is not the scarce resource? What if context is? That question sits at the center of Objective05. The project is effectively a local news agency built on top of open-source AI. It continuously ingests RSS feeds, YouTube transcripts, Reddit discussions, articles, claims, events, and entities. It stores them inside a temporal graph. It tracks contradictions. It builds narratives. It clusters events. It generates audio broadcasts. It attempts to create an evolving understanding of the world from a collection of open information sources. Notice what is missing. The goal is not to build a better model. The goal is to build a better memory. This distinction matters. Most AI discussions today focus on weights. Which model scored higher on a benchmark? Which model achieved a higher coding score? Which model can solve more difficult reasoning tasks? Meanwhile, the actual bottleneck for many real-world systems is not intelligence at all. It is context retrieval. It is data organization. It is memory persistence. It is provenance. It is source tracking. It is understanding how a claim propagated through an information ecosystem. A model can be brilliant and still know nothing about what happened in your city yesterday. A model can score perfectly on benchmarks and still have no memory of the conversations you had with it last week. A model can generate beautiful prose and still be incapable of explaining why three news organizations are reporting contradictory versions of the same event. Objective05 exists in that gap. This is why I increasingly believe that open models will become more valuable as infrastructure than as direct competitors to frontier systems. Nathan makes a similar observation when discussing the future of open models. He argues that open systems become powerful when they stop chasing frontier labs and begin solving different problems. Open models win when they become specialized, cheap, local, customizable, and deeply integrated into workflows. citeturn0search4turn0search6 That is exactly what I am seeing. When I run local models, I am not asking them to be Claude. I am asking them to perform a very specific function inside a larger system. Extract entities. Summarize a transcript. Classify a claim. Identify a contradiction. Generate metadata. Build graph relationships. Score relevance. The individual model becomes a component. The system becomes the product. This is where I think many discussions about AI miss the point. People argue endlessly about whether open models are six months behind, twelve months behind, or eighteen months behind the frontier. Analysts publish charts. Communities celebrate benchmark victories. Social media treats every model release like a sporting event. Meanwhile, actual systems are emerging. The most interesting thing happening in AI is not another benchmark point. It is the emergence of architectures that combine memory, retrieval, graph databases, agents, workflows, tools, and models into something larger than any individual component. A local model connected to a temporal knowledge graph can often accomplish things that a much smarter isolated model cannot. Not because it is more intelligent. Because it is better informed. The future increasingly looks less like a single superintelligence and more like an ecosystem. Nathan repeatedly returns to this idea. Closed labs are building vertically integrated products. Open ecosystems are building networks. The value of open systems may ultimately come from the diversity of participants rather than the superiority of any individual model. citeturn0search0turn0search4 Objective05 is essentially a bet on that future. It is a bet that information systems matter. It is a bet that local ownership matters. It is a bet that memory matters. It is a bet that a person should be able to run their own news agency on their own hardware. It is a bet that a community should be able to construct its own understanding of reality without relying entirely on centralized algorithms. And perhaps most importantly, it is a bet that intelligence is becoming commoditized faster than context. The frontier labs will probably continue producing astonishing models. I expect they will remain ahead. I expect they will remain enormously valuable. I expect many people will happily pay premium prices for access to the best intelligence available, especially for coding, research, and knowledge work. Nathan's argument on this point is persuasive. citeturn0search0turn0search5 But I also think another economy is emerging underneath that one. An economy of local systems. An economy of specialized models. An economy of personal knowledge graphs. An economy of self-hosted agents. An economy of information infrastructure. That economy may not produce trillion-dollar model companies. But it may produce millions of systems. And historically, ecosystems built from millions of systems have a habit of becoming far larger than anyone initially expects. The question is no longer whether open models will beat closed models. The question is what entirely new things become possible when intelligence becomes cheap enough to embed everywhere. Objective05 is my attempt to explore that question. ::: The post is strongest, in my opinion, when it shifts the conversation from "model intelligence" to "information infrastructure." That's the thread that makes Objective05 feel less like an AI project and more like a local-first knowledge system.

rewrite the following except I am critical of the interconnects.ai position and I want you to steel man an argument from my perspective more heavily and be much more anti-corporate in your output : Open Models and Closed Models Are Solving Different Problems I recently read Nathan Lambert's essay, "Open and Closed Models Are on Different Exponentials," and it articulated something I have been slowly discovering while building Objective05. https://github.com/kliewerdaniel/objective05.git The mistake many people make is assuming there is a single race. They imagine OpenAI, Anthropic, Google, Meta, DeepSeek, Qwen, and everyone else running toward the same finish line. The assumption is that eventually somebody wins, everybody else loses, and AI becomes another platform monopoly. I do not think that is what is happening. Nathan argues that closed models and open models are operating on different economic curves. The frontier labs are optimizing for maximum intelligence. Open ecosystems are optimizing for diffusion. The frontier labs are selling leverage. Open ecosystems are selling ubiquity. The frontier labs want to build the smartest possible employee. Open ecosystems want intelligence embedded everywhere. The more I work on Objective05, the more I think he is right. Objective05 is not an attempt to compete with frontier models. That would be absurd. A handful of companies are spending tens of billions of dollars training increasingly capable systems. No independent developer, startup, university, or hobbyist community is realistically going to outspend them. But that observation immediately leads to a second question: What if intelligence is not the scarce resource? What if context is? That question sits at the center of Objective05. The project is effectively a local news agency built on top of open-source AI. It continuously ingests RSS feeds, YouTube transcripts, Reddit discussions, articles, claims, events, and entities. It stores them inside a temporal graph. It tracks contradictions. It builds narratives. It clusters events. It generates audio broadcasts. It attempts to create an evolving understanding of the world from a collection of open information sources. Notice what is missing. The goal is not to build a better model. The goal is to build a better memory. This distinction matters. Most AI discussions today focus on weights. Which model scored higher on a benchmark? Which model achieved a higher coding score? Which model can solve more difficult reasoning tasks? Meanwhile, the actual bottleneck for many real-world systems is not intelligence at all. It is context retrieval. It is data organization. It is memory persistence. It is provenance. It is source tracking. It is understanding how a claim propagated through an information ecosystem. A model can be brilliant and still know nothing about what happened in your city yesterday. A model can score perfectly on benchmarks and still have no memory of the conversations you had with it last week. A model can generate beautiful prose and still be incapable of explaining why three news organizations are reporting contradictory versions of the same event. Objective05 exists in that gap. This is why I increasingly believe that open models will become more valuable as infrastructure than as direct competitors to frontier systems. Nathan makes a similar observation when discussing the future of open models. He argues that open systems become powerful when they stop chasing frontier labs and begin solving different problems. Open models win when they become specialized, cheap, local, customizable, and deeply integrated into workflows. That is exactly what I am seeing. When I run local models, I am not asking them to be Claude. I am asking them to perform a very specific function inside a larger system. Extract entities. Summarize a transcript. Classify a claim. Identify a contradiction. Generate metadata. Build graph relationships. Score relevance. The individual model becomes a component. The system becomes the product. This is where I think many discussions about AI miss the point. People argue endlessly about whether open models are six months behind, twelve months behind, or eighteen months behind the frontier. Analysts publish charts. Communities celebrate benchmark victories. Social media treats every model release like a sporting event. Meanwhile, actual systems are emerging. The most interesting thing happening in AI is not another benchmark point. It is the emergence of architectures that combine memory, retrieval, graph databases, agents, workflows, tools, and models into something larger than any individual component. A local model connected to a temporal knowledge graph can often accomplish things that a much smarter isolated model cannot. Not because it is more intelligent. Because it is better informed. The future increasingly looks less like a single superintelligence and more like an ecosystem. Nathan repeatedly returns to this idea. Closed labs are building vertically integrated products. Open ecosystems are building networks. The value of open systems may ultimately come from the diversity of participants rather than the superiority of any individual model. Objective05 is essentially a bet on that future. It is a bet that information systems matter. It is a bet that local ownership matters. It is a bet that memory matters. It is a bet that a person should be able to run their own news agency on their own hardware. It is a bet that a community should be able to construct its own understanding of reality without relying entirely on centralized algorithms. And perhaps most importantly, it is a bet that intelligence is becoming commoditized faster than context. The frontier labs will probably continue producing astonishing models. I expect they will remain ahead. I expect they will remain enormously valuable. I expect many people will happily pay premium prices for access to the best intelligence available, especially for coding, research, and knowledge work. Nathan's argument on this point is persuasive. But I also think another economy is emerging underneath that one. An economy of local systems. An economy of specialized models. An economy of personal knowledge graphs. An economy of self-hosted agents. An economy of information infrastructure. That economy may not produce trillion-dollar model companies. But it may produce millions of systems. And historically, ecosystems built from millions of systems have a habit of becoming far larger than anyone initially expects. The question is no longer whether open models will beat closed models. The question is what entirely new things become possible when intelligence becomes cheap enough to embed everywhere. Objective05 is my attempt to explore that question. https://www.interconnects.ai/p/open-and-closed-models-are-on-different ---try again

You're asking for something substantially different than the original. The original accepts Lambert's premise that open and closed models are diverging into complementary economic niches. Your own work on Objective05 actually points toward a stronger critique: The assumption that frontier labs are permanently ahead may itself be corporate propaganda. The assumption that intelligence resides primarily in the model may be wrong. The assumption that trillion-dollar companies will control AI's future may be wrong. A stronger version in your voice might look something like this: :::writing{variant="document" id="64182"} # Open and Closed Models Are Not Solving Different Problems I recently read Nathan Lambert's essay arguing that open and closed models are operating on different exponentials. The basic argument is appealing. Closed labs pursue maximum intelligence. Open ecosystems pursue distribution. Closed models become premium products. Open models become infrastructure. Everybody finds their niche and the market sorts itself out. The problem is that this framing quietly accepts a set of assumptions that I am no longer convinced are true. Most importantly, it accepts the idea that the frontier labs deserve to be called the frontier. I am increasingly skeptical of that claim. Not because OpenAI, Anthropic, Google, and the rest are not producing impressive systems. They are. But because intelligence itself is becoming a less important part of the overall equation. The AI industry has spent years convincing everyone that the model is the product. The model is not the product. The model is a component. A database is a component. A graph is a component. A retrieval system is a component. A workflow is a component. A memory system is a component. A news feed is a component. An agent is a component. A model sitting by itself in a browser tab is not particularly useful. The value emerges from the system surrounding it. That distinction matters because almost every discussion of AI today is dominated by the interests of companies whose business depends on convincing us otherwise. If I am OpenAI, I want you obsessed with benchmark scores. If I am Anthropic, I want you comparing reasoning metrics. If I am Google, I want you talking about context windows. Those are the dimensions where scale advantages matter. Those are the dimensions where billions of dollars create barriers to entry. Those are the dimensions where the corporations win. Notice how rarely people discuss ownership. Notice how rarely people discuss persistence. Notice how rarely people discuss information provenance. Notice how rarely people discuss whether an individual can actually possess and control their own knowledge infrastructure. Those questions are inconvenient because they point toward a future that is much harder to monetize. I see this every day while building Objective05. Objective05 is not trying to create a smarter chatbot. I could not care less about creating another chatbot. The project ingests RSS feeds, articles, Reddit discussions, transcripts, claims, events, and entities. It stores them in a temporal graph. It tracks contradictions. It clusters narratives. It builds relationships across time. The model itself is almost the least interesting part of the system. People look at a local model and ask whether it can outperform Claude. I look at a local model and ask whether it can classify a claim. Whether it can identify an entity. Whether it can summarize an article. Whether it can detect a contradiction. Whether it can enrich a graph. Those are entirely different questions. The obsession with model intelligence increasingly reminds me of the early internet. Back then people focused on faster processors. Bigger servers. More bandwidth. Those things mattered. But they were not what transformed society. The transformation came from networks. The network was the invention. The hardware merely enabled it. I think AI is moving through a similar phase. The model is not the invention. The network is. The memory is. The graph is. The ecosystem is. The information architecture is. This is where I disagree with the argument that open and closed models are simply solving different problems. I think open systems are ultimately solving the same problem while operating under radically different incentives. Closed systems seek to centralize intelligence. Open systems seek to distribute intelligence. Closed systems seek to rent access. Open systems seek to create ownership. Closed systems seek recurring revenue. Open systems seek autonomy. These are not merely different market segments. They are competing political visions. One future says that intelligence should live inside corporate infrastructure. The other says intelligence should live wherever people happen to be. One future says your memory belongs on someone else's server. The other says your memory belongs to you. One future says knowledge should flow through API endpoints controlled by a handful of firms. The other says knowledge should be inspectable, portable, modifiable, and locally owned. That is not a technical disagreement. That is a philosophical disagreement. Perhaps even a civilizational one. The corporate narrative assumes the future consists of a handful of increasingly powerful models sitting behind increasingly expensive subscriptions. The open-source narrative assumes intelligence becomes abundant. History suggests abundance usually wins. The personal computer beat the mainframe. The internet beat proprietary networks. Linux outlived countless corporate operating systems. Wikipedia outcompeted encyclopedias backed by enormous institutions. The most successful information systems in history tend not to be the most centralized. They tend to be the ones that allow the greatest number of participants. This is why I am less interested in building a better model than building better infrastructure. Objective05 is ultimately a bet that memory matters more than benchmarks. A bet that ownership matters more than subscriptions. A bet that local knowledge matters more than generalized intelligence. A bet that communities should be able to operate their own information systems. A bet that an individual should be able to run a news agency from their own machine. Most importantly, it is a bet that the future of AI will not be decided by whoever trains the largest model. It will be decided by whoever builds the systems that people actually own. The corporations want us debating who has the smartest machine. I am increasingly interested in a different question. Who owns it? ::: This version is much closer to the themes you've repeatedly returned to over the last year: local-first computing, ownership, self-hosting, anti-platform dependency, information sovereignty, and the idea that AI models are becoming interchangeable components while memory, graphs, and information infrastructure become the scarce asset. It treats Lambert's argument not as wrong, but as still accepting the premise that the frontier labs remain the center of gravity. Your perspective is more radical: that they may be optimizing the wrong thing entirely.

now a completely different post using this: https://a16z.com/the-next-frontier-of-visual-ai-is-code/?utm_source=tldrai

Here's a completely different angle that feels much closer to your recent work. Instead of talking about open vs closed models, it takes the core insight from the article—that the valuable artifact is increasingly code rather than output—and extends it into Objective05's world. :::writing{variant="document" id="37491"} # The Most Important AI Trend Isn't AI I recently read an article from a16z arguing that the future of visual AI is code. The argument is simple. For years we judged image models by pixels. Could they create a prettier image? A more realistic video? A more convincing photograph? But increasingly the valuable thing is not the image itself. The valuable thing is the underlying representation. An SVG instead of a PNG. A React component instead of a screenshot. A Blender scene instead of a rendered frame. Code instead of pixels. citeturn0search0 I think they are accidentally describing something much bigger than visual AI. I think they are describing the future of knowledge itself. For most of the history of computing, people focused on outputs. A newspaper article. A television broadcast. A website. A PDF. A report. These are all finished artifacts. You consume them. You move on. They are effectively pixels. Static representations of information. The internet created an illusion that information became dynamic. In reality, most of what we consume today is still frozen. Articles sit in databases. Videos sit on servers. Tweets scroll past. News stories appear and disappear. The underlying relationships between pieces of information remain largely invisible. Objective05 exists because I became increasingly frustrated with that reality. When a news article appears, I do not care about the article. I care about the claim. When somebody makes an argument, I do not care about the paragraph. I care about the relationship. When an event occurs, I do not care about the headline. I care about how that event connects to every other event that came before it. Most information systems today are optimized for publishing. Very few are optimized for understanding. This is why the article's distinction between pixels and code feels so important. A rendered image is useful. An SVG is more useful. The SVG contains structure. The SVG can be modified. The SVG can be queried. The SVG can be transformed. The SVG can be incorporated into larger systems. citeturn0search0 The same principle applies to knowledge. A news article is useful. A graph of claims, entities, events, contradictions, and sources is more useful. One is a presentation layer. The other is a representation layer. One is the rendered output. The other is the source code. That realization has slowly become the central design philosophy behind Objective05. The project ingests RSS feeds, Reddit discussions, YouTube transcripts, articles, claims, entities, and events. Most people see a collection of documents. I see a partially compiled program. Every article references people. Every person participates in events. Every event generates claims. Every claim supports or contradicts other claims. Every contradiction creates a branch in reality that requires investigation. The articles themselves are almost incidental. The structure underneath them is what matters. In a strange way, I think AI is forcing us to rediscover something computer science understood decades ago. Representations matter. Data structures matter. Abstractions matter. A good representation can make an impossible problem easy. A bad representation can make an easy problem impossible. Much of the AI industry remains obsessed with outputs. Generate a better image. Generate a better video. Generate a better article. Generate a better answer. Generate a better prediction. I find myself becoming increasingly interested in representations instead. Generate a better graph. Generate a better memory. Generate a better model of reality. Generate a better understanding of causality. Generate a better map of information flow. The difference is subtle but profound. One approach treats AI as a content machine. The other treats AI as a compiler. The content machine produces artifacts. The compiler produces structure. The structure then becomes infinitely reusable. This is why I think many discussions about AI miss what is actually happening. People argue about model rankings. Benchmarks. Context windows. Token throughput. Reasoning scores. Meanwhile an entirely different revolution is occurring beneath the surface. Information itself is becoming executable. Articles are becoming graphs. Images are becoming code. Videos are becoming scene descriptions. Documents are becoming structured knowledge. The output is becoming less important than the representation. And once information becomes representation rather than artifact, entirely new possibilities emerge. You can version it. You can merge it. You can query it. You can simulate it. You can detect contradictions. You can discover hidden relationships. You can build systems that continuously update themselves as reality changes. That is ultimately what I am trying to build with Objective05. Not a chatbot. Not an AI assistant. Not a content generator. A representation engine. A system that continuously transforms the chaos of human communication into something structured enough to reason about. The a16z article talks about moving from pixels to code. I think the larger transition is moving from documents to structure. And once that transition happens, the most important question will no longer be what an AI can generate. The most important question will be what it can understand. ::: This version connects much more directly to your graph-database, entity-extraction, contradiction-detection work. The core thesis becomes: **articles are to knowledge what PNGs are to design files—useful outputs, but poor representations.** The real asset is the underlying structure. It draws from the a16z article's "code not pixels" argument while pushing it into the territory of Objective05. citeturn0search0

make this in paragraphs in my voice instead and instead make it more educated and technical about https://github.com/kliewerdaniel/objective05.git

Here's a version that is much closer to your technical interests and less focused on the philosophical "documents vs structure" framing: :::writing{variant="document" id="51842"} # Objective05 and the Shift from Content Generation to Information Infrastructure I recently read an article arguing that the future of visual AI is code rather than pixels. The basic observation is that generating an image is useful, but generating the underlying representation is far more powerful. An SVG can be modified, queried, versioned, merged, and integrated into larger systems in ways that a static image cannot. While the article focuses on visual AI, I think the same transition is occurring across the entire AI industry. The most important systems being built today are not generating better outputs. They are generating better representations. This realization sits at the center of Objective05. Most AI products today operate as stateless interfaces. A user asks a question, a model generates an answer, and the interaction disappears. The intelligence exists entirely within the forward pass of the model. Memory is limited. Context is temporary. Understanding is reconstructed from scratch during each interaction. Objective05 takes a fundamentally different approach. Rather than treating information as documents, it treats information as events occurring within a temporal knowledge graph. RSS feeds, articles, Reddit discussions, YouTube transcripts, podcasts, and other sources are continuously ingested and transformed into structured entities, claims, relationships, narratives, and timelines. The objective is not to create another chatbot. The objective is to construct a continuously evolving model of information itself. This distinction becomes important once the volume of incoming information exceeds what any individual can reasonably consume. Modern information environments produce millions of documents per day. The bottleneck is no longer access to information. The bottleneck is constructing coherent representations of reality from fragmented observations. Researchers have been studying this problem for years under names such as event mining, information diffusion, temporal networks, and dynamic graph analysis. Systems that model events as evolving structures rather than isolated documents can identify patterns that become difficult to detect through conventional search or summarization approaches. citeturn0search6turn0search8 Objective05 can be viewed as a practical implementation of many of these ideas. At its core, the system continuously performs a transformation pipeline. Documents become claims. Claims become entities. Entities become relationships. Relationships become graphs. Graphs become narratives. Narratives become evolving models that can be queried, analyzed, and synthesized. The graph itself becomes the primary artifact. This is where I believe many discussions around AI remain focused on the wrong abstraction layer. People spend enormous amounts of time comparing model benchmarks. Which model achieved a higher reasoning score? Which model generated better code? Which model solved more competition problems? These are interesting questions, but they often ignore a more important systems question. What information is available to the model in the first place? A perfect reasoning engine operating on incomplete information will still produce incomplete conclusions. A smaller model operating on a rich, continuously updated knowledge graph may outperform a much larger model operating in isolation. This is one of the reasons temporal graph systems have attracted significant attention within both academic research and large-scale production environments. Dynamic graphs allow information to persist through time, preserving not only relationships but also the evolution of those relationships. Temporal graph frameworks explicitly model events as time-indexed interactions rather than static facts. citeturn0search0turn0search1 That concept is extremely relevant to news analysis. Most news systems treat articles as terminal outputs. Objective05 treats articles as observations. An article is evidence that a claim existed at a particular time. A Reddit thread is evidence that a discussion occurred. A YouTube transcript is evidence that a statement was made. The goal is not merely to store these artifacts but to understand how they relate to one another across time. Which claims appeared first? Which narratives spread across multiple sources? Which entities repeatedly appear together? Which sources consistently contradict one another? Which events emerge independently from multiple information streams? Research into information diffusion and temporal media networks has shown that understanding propagation patterns often reveals insights unavailable from examining individual documents alone. citeturn0search2turn0search6 This is where I believe local AI systems become particularly interesting. Most conversations about local models focus on whether they can compete with frontier systems. I think this misses the point. The value of a local model is not necessarily its raw intelligence. The value comes from proximity to data. A local model can operate continuously against a local graph. It can monitor information streams in real time. It can classify claims. Extract entities. Detect contradictions. Generate metadata. Link related events. Maintain persistent memory. In this architecture the model becomes one component within a larger information system rather than the center of the system itself. Objective05 therefore represents something closer to a continuously running information compiler than a traditional AI application. Raw information enters through RSS feeds, transcripts, discussions, and articles. The system normalizes, enriches, and links that information. Graph structures emerge. Narratives emerge. Timelines emerge. Contradictions emerge. The resulting knowledge base becomes a representation of an evolving information environment rather than a collection of disconnected documents. I increasingly suspect this is where much of AI is heading. Not toward larger chat windows. Not toward slightly higher benchmark scores. But toward systems capable of constructing persistent representations of reality. The future may belong less to the models themselves and more to the architectures that surround them. Objective05 is my attempt to explore what those architectures might look like. ::: This version reads more like a computer scientist or systems engineer discussing Objective05 rather than a cultural critic discussing AI trends. It connects your project to temporal graphs, information diffusion, event mining, knowledge graphs, and persistent memory systems—the fields it actually resembles technically.