"If we carefully closed the right feedback loops, could we construct a creative flywheel that generates finished works almost by accident, through a stream-of-consciousness process?"

"Knowledge gardening is one kind of system for generating self-organizing ideas. As a kind of self-organizing system, it can’t happen without some kind of feedback loop."

"This is what broken feedback loops feel like. Everything is feed-forward. There are no opportunities in the interface to stumble back over your notes, to recurse back over them. You’re continually adding energy to the system, but none of it is returning back."

"Knowledge gardening, on the other hand, is the emergent result of constructing a feedback system. I recurse over scratch notes, revise them, add to them, refactor them, and combine them with other ideas to form new ideas."

"The core game mechanic of Zettelkasten is to file your note some place where you would want to stumble over it again."

"Like Zettelkasten’s game mechanic, this search-or-create mechanic closes a feedback loop. Every time you enter a new idea, you’re recursing over old ideas. Often you find yourself editing, or refactoring an old note, instead of creating a new note. Over time, these microinteractions add up, generating knowledge from the bottom-up."

"Are social graphs even a moat? Perhaps more than data, but TikTok proves you don’t need a dense social graph to serve up interesting content."

"What information consumes is rather obvious: it consumes the attention of its recipients. Hence a wealth of information creates a poverty of attention, and a need to allocate that attention efficiently among the overabundance of information sources that might consume it. (Herbert A. Simon)"

"Technology for producing and distributing information is useless without some way to locate, filter, organize and summarize it. A new profession of “information managers” will have to combine the skills of computer scientists, librarians, publishers and database experts to help us discover and manage information. These human agents will work with software agents that specialize in manipulating information-offspring of indexing programs such as Archie, Veronica and various World Wide Web crawlers that aid Internet navigators today."

"Hal Varian, by the way, is the person who designed the Google ad auction."

"So, the condition of superabundance creates a need for aggregation. LLMs will amplify that abundance. It’s a good bet that this strengthens the strategic importance of aggregation."

"I think we will need to sign everything to signify its validity. When I say sign, I am thinking cryptographically signed, like you sign a transaction in your web3 wallet."

"I think this may break the web’s security model. The web has a fundamentally feudal structure. It conceptualizes security as a castle wall around the server. The server controls the keys."

"We have to sign everything. We need to start thinking about security the way crypto thinks about security. Not your keys, not your data."

"By contrast, user-owned keys enable user-owned data. Self-sovereign keys give apps one less chokepoint by which to lock you in."

"Perhaps this might result in faster emergence of aggregators and also faster collapse. A hotter innovation loop. New aggregators could rapidly emerge and compete, without having to contend with incumbent network effect."

"LLM vendors seem more like a traditional industrial monopoly. Like an industrial monopoly, the moat here is capital cost—the cost of gathering the data, and the cost of training the model."

"But industrial monopolies rely on the means of production remaining scarce. Is this a safe assumption in software? Is it a durable moat?"

"I suspect there will be a lot of pressure to commoditize LLMs. Why? LLMs are the complement to many consumer products, and smart companies try to commoditize their compliments:"

"Open source models may also be at a competitive advantage because they enable permissionless innovation."

"The state of the art in AI was all teddy bears and cowboy astronauts—inoffensive images generated by product managers—until Stable Diffusion was released."

"One good-enough open source model, and we saw a proliferation of product concepts and demos."

"Permissionless models have more variety, in the cybernetic sense."

"The structure of a network itself, more than the regulations which govern its use, significantly determines what people can and cannot do."

"Centralized networks tilt toward centralized governance. The money and power are centralized, so governance is too. Benevolent monarchy at best, totalitarianism at worst."

"Centralization is often necessary for new product innovation. Innovations often start out vertically integrated in order to line up all the pieces in the stack."

"Bottlenecked by bureaucracy. We can think of the central bureaucracy as a CPU who’s job is to compute policy decisions for the rest of the network. The complexity of the problems that can be dealt with is limited by the complexity of the central bureaucracy."

"If a system is to be stable, the number of states of its control mechanism must be greater than or equal to the number of states in the system being controlled."

"Another way to see this is that a centralized bureaucracy does not have requisite variety to govern organic growth. The edge is more expressive than the center, and changes more quickly. This forces the center to limit the variety at the edge in order to maintain control of the system."

"Few feedback mechanisms."

"Because people at the edge don’t have power, it is difficult for them to create feedback mechanisms to pass information back up to center."

"Reality debt builds up between center and edge."

"While centralization first evolves for reasons of efficiency, the purpose of the resulting hierarchy soon shifts toward its own perpetuation."

"Philosopher Charles Taylor refers to this kind of structure as a Direct-Access Society. In a Direct-Access Society, people are conceived of as individuals, disembedded from social hierarchies, free economic, or voluntarist agents."

"We could say P2P centralizes the system around the protocol, in an attempt to distribute something else (control, resources, or power). This means governance has to be encoded within the protocol itself."

"Federated, or decentralized (B), networks tilt toward a patchwork of private and public shared spaces. They are a hybrid model, where many small-to-mid-sized hubs agree to cooperate together using shared protocols."

"This grants federation more variety than centralized one-size-fits-all networks. It also grants federation more variety than p2p networks, since a protocol can never have requisite variety to encompass the complexity of human nature."

"Abuse of power can happen at the fed layer."

"In fact, centralization emerges organically in all mature ecosystems."

"Centralization is efficient."

"Centralization has economies of scale."

"So the more connections a node gets, the more it gets. This is called preferential attachment, and it happens around any kind of more-gets-more Matthew Effect: connections, capital, power. Preferential attachment results in networks with power law distributions."

"Federated networks are always in danger of collapsing into centralized networks for this reason."

"Evolved systems end up being tapestries of centralization, decentralization, and federation, for these reasons."

"A network of 150 friends has 22,350 relationships to keep track of. That’s quite a few, and the number grows exponentially with the number of friends: (n * (n-1))."

"The bi-directional network scaling problem. Beyond 150, we start to run up against cognitive limits. This limit is called Dunbar’s Number."

"Beyond 150, you start to run into problems. You can’t keep track of all the players, so free-riders get away with it. Sociopathic behaviors become personally advantageous. This isn’t just true for hunter-gatherers."

"The ugly truth is that we believe in connecting people so deeply that anything that allows us to connect more people more often is *de facto* good."

"The internet makes it possible to connect everybody to everybody. Social media leveraged this capability to rapidly bootstrap massive network effects."

"We’re way past our Dunbar limit. In these conditions, trust is nonexistent, sociopathic behaviors are personally beneficial. There is little incentive to cooperate."

"In the meantime, what if we scaled down? We can already see this happening, people retreating to group chats, Discords, Slacks. Smaller, high-trust communities."

"Subtext is line-oriented. Each line in the file is treated as a discrete block of content."

"The right mental analogy for Subtext is not the page. It is the the index card."

"Subtext represents block-oriented documents as line-oriented markup."

"Blocks are thought legos. A block-oriented document is composable (and decomposable). You can break it apart into component blocks, filter it down to blocks of a particular type, merge documents, pluck out blocks, link to specific blocks, etc."

"Linear block-oriented documents are like shipping containers for discrete thoughts. Because blocks are structurally uniform, they can be automatically moved around and reorganized. Software can split, join, and merge documents easily and effectively, because the document structure is simple."

"The plan is to have Subconscious display these links as transclusions. Rather than linked words in text, imagine something more like a quote tweet… Links to images display as literal images, links to videos display as playable videos with playback controls, links to documents display some or all of the content inside of the linked document. This lets you compose hypertext documents from many smaller documents."

"Big documents can be composed by linking to small documents."

"However, over time, I noticed that my markup needs for note-taking were different from my markup needs for publishing."

"My note-taking style organically converged on a tiny subset of Markdown's features: text, links, lists, quotes, and one level of heading. To have more may be useful for publishing, but is often overkill for note-taking."

"I started to run into limitations with Markdown and HTML. As a complex publishing format, it is unclear how to meaningfully decompose or merge Markdown/HTML documents."

"When you combine documents, heading levels may need to be changed, lists may need to be flattened or nested. Because the document format is complex, foreknowledge of the meaning of the document is necessary to make meaningful changes."

"Subtext is an attempt to resolve the problem by radically simplifying it. Paradoxically, by limiting the format to a flat list of blocks, we radically expand what software can do with it."

"When you train a deep learning system, it builds up a a latent space. A latent space is a hyperdimensional space where things that are similar along some dimension are near to each other along that dimension."

"Latent space is like a map you can use to correlate things with other things, along many dimensions."

"What if we imagined ideas as a kind of hyper-dimensional latent space?"

"Imagine your notes as a stack of papers. Next to that stack of papers, you have a long red string, with a needle at one end. To perform a search, you leaf through each of the pages. Whenever you find a match for your query, you poke the needle through, right where the match exists on the page, and you thread the page onto the string. You’re methodical, so you order the pages on the string by relevance, taking into account how close the match is, and where it appears on the page."

"Now, stretch out the red string. You’ve arranged your ideas along one dimension of latent idea space. Imagine doing this for all possible dimensions at once."

"Any sufficiently advanced search is indistinguishable from a hyperlink."

"When a search becomes extremely specific, it functions like a coordinate to a specific point in latent idea space."

"Where a book offers one narrative linearization of a topic, and hypertext offers a static graph of relationships, we might see search as a dynamic graph."

"Search is a signal of intent. It reveals the useful dimensions within latent idea space. By asking the question, we are signaling which possible connections, in the space of all possible connections, matter to us."

"Questions are also knowledge, and can be used to construct both answers, and questions."

"If you've ever maintained a Wiki, you've probably noticed that there is a lot of refactoring involved. Ideas are written down in one place, then rewritten, moved, titles changed, links redirected, pages split and merged."

"Hypertext wants to be refactored. This is a feature of hypertext, not a bug. Through constant refactoring, knowledge in a hypertext network evolves to find the right packets for a given domain, where one packet = one idea."

"Refactoring lets us construct both our question and our answer through organic exploration."

"The kind of map you construct depends upon what you’re trying to see."

"Factoring out documents. Notes often start out as a jumble of ideas. As a note evolves, separate ideas are factored out into their own notes."

"Note that if we look at this in terms of Capture, Organize, Synthesize, many of these actions are capture and organize. Capture and organize are both modular, rather than holistic actions."

"Rewriting may be trickier, because it requires holistic synthesis that spans the boundaries of any single idea, paragraph, or page."

"If we want to augment or automate refactoring, we need a way to structure text into logical blocks that can be factored out, merged, added, rearranged, removed. And so… Subtext."

"What if we introduced the minimum amount of structure for working with text? Something simple for people, simple for computers, and meaningful for both?"

"Is the thinking in the pencil? The cybernetic resolution to this paradox is to shift our focus away from static objects—pencil, piano, person—and toward the flow of information between those objects."

"The pencil and the person make up a feedback loop. The piano and the person make up a feedback loop. Together they make up an extended-self system that can do things which neither could have done alone."

"It emerges from the feedback between the piano and me."

"Familiar tools are represented by the brain as an extension of the body."

"Like the cybernetic shift from object to message, the concept of the extended phenotype is a Copernican shift. It resolves a paradox by decentering the organism."

"For Dawkins, who developed the idea, it is not the organism that is unit of selection, it is the selfish gene. The rest (including me) is a just a vehicle, and the boundaries of that vehicle are incidental. One self-system can freely bleed into another."

"To quote Brian Eno, “the great benefit of computer sequencers is that they remove the issue of skill, and replace it with the issue of judgement”."

"Each distinct phase requires a different creative strategy:"

"Capture as much as possible without judgment."

"Organize to give some first-pass structure using rough heuristics. Card sorting, clustering, filing, arranging, etc."

"Synthesize new meaning."

"Why do we expect ourselves to create good ideas from nothing (synthesize)? It’s much easier to generate ideas when you have lots of material (capture) clustered by themes and relationships (organize)"

"what if instead of fighting a broken system through force of will, we constructed a new system—a cybernetic system—designed to amplify capture-organize-synthesize? If we carefully closed the right feedback loops, could we construct a creative flywheel that generates finished works almost by accident, through a stream-of-consciousness process?"

"Self-organization requires a mechanism of composition"

"Self-organization happens through composition. Composition is combining things with other things to create new things."

"Alphabets tend to follow Gall’s Law: simple alphabets produce complex behaviors, complex alphabets produce stupid behaviors."

"Universal mechanisms of composition maximize expression"

"Evolution is composition with memory"

"Evolution emerges in any system with:"

"Mutation"

"Heredity"

"Selection"

"When we write, we flatten the cloud of associated ideas in our head into a linearized subset (lossy). The reader then unflattens this linearized subset into their own cloud of associated ideas (lossy). Each lossy step is an opportunity for mutations in understanding to emerge. Useful mutations are remembered (selection) and shared (heredity). And so... mutation, heredity, selection. Ideas evolve!"

"If we want to generate self-organizing ideas, perhaps what we want is less like prose, and more like an index card, where 1 card = 1 idea."

"Not formatting, but the smallest amount of structure to unbundle text into sub-atomic units, so it can be composed and recomposed."

"Notes are conversations across time."

"It is a shift in perspective that can be grounded in cybernetics theory, particularly Gordon Pask’s Conversation Theory, and it yields a number of surprising design insights."

"Conversation Theory takes another view. It sees knowledge as conversational. Knowledge exists subjectively in our minds, and is constructed through conversation with others."

"As we converse, each of us constructs our own subjective model. I serialize my subjective understanding into a message. You receive this message, interpret it, update your subjective understanding, and then respond by serializing your own message, which I receive..."

"The loop stops when participants mutually decide they have converged toward a close enough shared understanding."

"Each participant has a different subjective model of the world. We can draw from our differing subjective understandings to knock conversation out of convergence."

"The crucial requirement is for the self to allow the other to “speak back” and to accommodate the unexpected so that self affects other, and other affects self. Avoiding requisite variety, both get partially “out of control” in a mix of positive and negative feedback, thus conversing along non-determinable trajectories to arrive and previously unknown destinations."

"Conversations don’t just happen between people."

"A conversation can happen between yourself and yourself, across time, through the notes your past self took for your future self."

"Cybernetics shifts our focus away from objects, and toward the flow of information between objects. Design through the lens of Conversation Theory is not about features and screens. Design is the making and breaking of cybernetic feedback loops."

"They know feedback loops are nonlinear, and force of will is only linear."

"We can construct conversational feedback loops that program creativity, or garden ideas from the bottom-up, or evolve ideas spontaneously."

"I think of this kind of process as knowledge gardening. You collect and plant idea seeds, returning periodically to water and weed. After many days, you find they have grown into a rough outline, a hypothesis, or even a complete work while you were away."

"Here’s the thing: the feedback loops of most note-taking apps are broken, or non-existent."

"Do you ever get the feeling you’re throwing your notes into the void?"

"Most of them are scratch notes—rough, incomplete, touched once, then forgotten."

"This is what broken feedback loops feel like."

"There are no opportunities in the interface to stumble back over your notes, to recurse back over them."

"Zettelkasten is an old and surprisingly effective paper system for idea gardening."

"It’s an index-card-based system, and that means you have to do a lot of manual note-card filing. All that filing can be quite a bit of work, but Zettelkasten leans into this limitation to construct a feedback loop."

"Notational Velocity was a tiny note-taking tool built around a single idea: search or create."

"Gardening feedback loops is tricky."

"They’re spread over time, not space."

"Like Zettelkasten’s game mechanic, this search-or-create mechanic closes a feedback loop. Every time you enter a new idea, you’re recursing over old ideas."

"Over time, these microinteractions add up, generating knowledge from the bottom-up."

"The gist is that I’m building a creative oracle that helps provoke ideas. It also captures those ideas, and remixes and resurfaces them, provoking more ideas... in a feedback loop that powers a flywheel of divergent creative thinking."

"Building convivial tools means breaking with much of the standard SaaS/Aggregator playbook, where a company owns and controls the software and the data."

"What if we instead had a small tool that was personal, multiplayer, distributed, evolvable?"

"Personal:"

"Multiplayer: Personal doesn’t have to mean solitary. One of the most interesting ways of thinking is thinking together. “Knowledge production is a group activity, not an individual one.” (Engelbart)."

"Evolvable"

"Distributed:"

"Hypertext resists the single linear narrative."

"Thoughts, also, resist the single linear narrative. They exist as something more like a hyperdimensional cloud of associations."

"When we talk together, we flatten our N-dimensional thoughts into 1D linear narratives, in order to fit them through the 1D bottleneck of words. When we draw or sketch, we flatten our N-dimensional thoughts into 2D images in order to fit them through the 2D bottleneck of sight."

"Literary montage was Benjamin’s attempt to get beyond the single perspective of the linear narrative, to see from many angles at once."

"I create accidental montage while taking notes. Montage is almost an intermediate representation between thought and narrative."

"Subconscious is a box of thought legos. It helps you assemble those legos in multiple ways, to provoke new meanings."

"In an important sense there are no subjects at all; there is only all knowledge, since the cross-connections among the myriad topics of this world simply cannot be divided up neatly. Hypertext at last offers the possibility of representing and exploring it all without carving it up destructively."

"Files act as hubs around which interoperability can emerge."

"How and where these protocols emerge is contingent on network effects, incentives, path dependency, and other strategic factors, but I think I see some themes."

"The early open networks: TCP/IP, HTTP, IMAP. These often came out of government-funded labs that weren’t bound by profit-seeking incentives. These protocols attempted to solve practical problems with layerable APIs that could build on each other, and support an open-ended set of use-cases."

"Low-level primitives: images, fonts, sound. Some mix of licensing models, patent pools, and open standards."

"Commercial exchange: Schema.org, sitemap.xml. Protocols in this zone emerge to commoditize their compliments."

"n * (n-1). That’s our directional graph equation again, but this time the network effect is on our side."

"Expect the output of every program to become the input to another, as yet unknown, program."

"The Unix Philosophy got this right from the beginning:"