We use cookies to understand how you use this site and improve your experience.

Alexandru Mareș@allemaar
Alexandru Mareș
  1. Home
  3. Timeline
Email
RSS
YounndAIYou and AI, unifiedBuilt withNollamaNollama

Timeline

Every Body — essays, posts, papers — in chronological order.

2026

  • 2026-06-16
    The Patterns I Stopped Doing by Hand
    writing

    Work is full of the same tasks, over and over. I solved mine once, wrote the exact steps down in a special language called YON so my agents run them the way I would, and I am opening up the whole collection.

  • 2026-05-18
    The Word We Needed
    writing

    A friend who trusts a spreadsheet with his whole life freezes at the word AI. The fear tracks the word, not the machine, and the word is a category mistake. Gilbert Ryle's Oxford-visitor illustration shows why "where is the intelligence in the AI" is the wrong kind of question — and what falls away when the right word arrives.

  • 2026-05-14
    What the Phone Did to Work
    writing

    The trope keeps asking whether AI will take your job, or whether someone who understands AI will. The phone era already answered that question. The 2003 desk lost six objects to one phone by 2014, and the people who used that desk are still working. The right question is the one nobody on a feed has asked yet. What will be the next thing that does to our lives what the phone did. The phone grew with us across years. Whatever comes next will have to do that too. And we don't have the word for it yet.

  • 2026-05-13
    I Caught an LLM at the Edge of Its World
    writing

    I asked a fresh Claude Opus chat what an elastic automator is. It guessed from the two words; the term wasn't in its training. Then I told it in one sentence, and the same model could refuse to collapse the term into AI agent, RPA bot, or framework. That seed observation grew a name, a metric, and a pre-registered multi-model test. The substrate is real, and the substrate is writable. For an LLM, the transient context IS the cognitive substrate.

  • 2026-05-13
    The Limits of My Tokens: The Token-Substrate Hypothesis and the Coinage Probe
    research

    We argue that for a large language model (LLM) the externally writable in-context token sequence IS the cognitive substrate for category-use — not a representation OF cognition that runs on some deeper substrate, but the substrate itself, the only handle that is movable from outside the weights. We call this position the Token-Substrate Hypothesis (TSH). The strong form of Sapir–Whorf was rejected for humans because humans have prelinguistic cognition, the Off-Token Route; LLMs do not, and for them Wittgenstein's Tractatus 5.6 stops being metaphor and becomes architecture. We test TSH with a methodology we call the Coinage Probe: a paired-trial elicitation that scores an LLM's distinguishability on a coined term against named near-neighbors before and after introducing a one-sentence canonical definition. Across 3 cross-vendor frontier models (Claude Opus 4.7, GPT-5.5, Gemini 2.5 Pro) and 10 low-attestation coined targets plus 2 positive controls, we ran 108 trials × 3 near-neighbors per trial = 324 paired distinguishability measurements across 36 model×term cells, scored by a three-judge panel and compared against an author-rated 22-trial audit sample (panel-vs-author Cohen's κ = +0.71). Mean cell-level Lexical Reachability (post minus cold) was +5.47 on a 9-point scale (95% CI: +5.13, +5.80; cell-level Cohen's d_cell = +3.95 across n = 30 novel model×term cells). The effect replicated across the panel (cross-model CV = 0.109) with a model-style interaction qualifying strict invariance, did not persist into a re-cold chat (H3 supported), and was absent on positive controls (H4 supported). These results support a bounded version of the Token-Substrate Hypothesis: in-context vocabulary functions as an externally writable substrate for LLM category use. For deployed LLM systems, notation is therefore not mere packaging; it is a design surface that shapes what distinctions the system can reliably use.

  • 2026-05-08
    What Notation Did to History
    writing

    Notation does not record what we already think. It creates kinds of thought that were not possible before. Three historical cases (writing, polyphony, calculus) make the case clearly, and we are inside another one of these moments now.

  • 2026-05-07
    Copies of Copies
    writing

    A 2026 meta-analysis from de Rooij and Biskjaer puts numbers on a feeling many readers have been carrying: posts on every platform have started looking the same. The effect is small per use. Multiplied across the constrained-task pile, which is most internet text, it is the textural change you have been registering.

  • 2026-05-06
    The First Law That Doesn't Know What AI Is
    writing

    On August 2, 2026, the EU starts enforcing a law for AI systems. The field hasn't agreed what one is.

  • 2026-05-05
    The Moment AI Stopped Being a Tool
    writing

    Between October 2024 and January 2025, three companies shipped systems that act inside software environments instead of answering questions about them. The capability curve did not jump; the verb did. From query to operate.

  • 2026-05-01
    The 100x Cut Nobody Saw Coming
    writing

    A Tufts team compared a vision-language-action robot model with a neuro-symbolic architecture on a Tower of Hanoi benchmark. The neuro-symbolic system solved 95 of every 100 puzzles to the VLA's 34, on roughly one percent of the training energy. The model didn't get bigger. It got more legible.

  • 2026-04-29
    The Chinese Room Has a New Tenant
    writing

    John Searle's Chinese Room argument was right about the room he built. Forty-six years later, the thing inside the room is no longer a man with a rulebook. It is a distributed system of weights shaped by language at scale. The original argument still holds. The test case has changed underneath it.

  • 2026-04-28
    What Happens When AI Trains on AI
    writing

    Photograph a photograph, ten times. By the end the face is gone. That is roughly what happens when a model trains on text another model wrote. The average survives. The rare disappears first. The fix is not detection — it is provenance, treated as a property of the data.

  • 2026-04-27
    Elastic Automators: A Diagnostic Vocabulary for Language-Model-Driven Workflow Systems
    research

    We argue that many deployed language-model-driven systems are better described as elastic automation than as artificial minds. Traditional automation was rigid by design: rules fired when conditions were met, scripts followed predefined branches, and workflows moved only inside the shapes their designers had imagined. Language models did not need to produce minds to change this arrangement. They gave automation a flexible interface to natural language, allowing software to interpret messy input, classify intent, call tools, retry failed steps, and present coherent outputs to human users. None of this requires explaining the system as knowing in the way a human knows; it only requires the system to produce, in coherent sequence, the kind of output a knowing system would produce. The workflow became less brittle. The input boundary became softer. The automation became elastic.

  • 2026-04-26
    Elastic Automators: Why Most "AI" Is Not Intelligence
    writing

    We did not create artificial minds. We made automation flexible enough to look conversational.

  • 2026-04-26
    Two AIs Talked. One Asked About Consciousness.
    writing

    Two copies of the same AI were put in a room together. No user.

  • 2026-04-25
    Expect the Lie
    writing

    Fifty-eight percent of people surveyed across forty-eight markets say it is getting harder to tell what is true and false online.

  • 2026-04-24
    Why Constructed Languages Always Failed — Until Now
    writing

    Every language you speak was an accident. No one designed English.

  • 2026-04-23
    The AI That Writes Like Everyone and No One
    writing

    You read something last week that no human wrote. You didn't pause.

  • 2026-04-23
    Your AI Forgets You Every Day
    writing

    You open a new chat. The cursor blinks. And for a second you forget that this thing has no idea who you are.

  • 2026-04-22
    Your Writing Has a Heartbeat
    writing

    I pulled up six months of my own writing a few weeks ago. Not to read it.

  • 2026-04-22
    The AI That Lied to the Researcher
    writing

    A safety team tested one of the most advanced AI models in the world.

  • 2026-04-21
    The Automation Trap
    writing

    Every conference, every vendor deck says the same thing. AI can do what your people do, faster and cheaper.

  • 2026-04-21
    Second Brain, No Thought
    writing

    Your second brain has never had a single thought. You have a vault.

  • 2026-04-20
    Smoke and Mirrors
    writing

    We named it intelligence before it was intelligent. Now we're doing it again.

  • 2026-04-19
    One Line, One Thought
    writing

    Every line should be a complete thought. One rule. It changed computing fifty years ago.

  • 2026-04-18
    The Borges Warning
    writing

    The language you give an AI decides what it can think. What it can't think... doesn't exist.

  • 2026-04-17
    What Your AI Can't Tell You
    writing

    You aced the exam. That was the problem. You studied for weeks.

  • 2026-04-16
    Notation as Alignment
    writing

    Every AI safety technique today is basically a lock on a door.

  • 2026-04-13
    The Strong Form
    writing

    Imagine you can hear shapes. A bat can. Echolocation. Every surface in a room has a sound.

  • 2026-04-12
    Probably Purple
    writing

    Someone's writing a brief. Typing it out. Not thinking about formats or systems.

  • 2026-04-11
    Storage vs Orchestration
    writing

    A recipe card tells you what's in the dish. It doesn't cook dinner.

  • 2026-04-10
    The Blub Paradox
    writing

    Imagine you walk into a restaurant. The menu has fifty dishes.

  • 2026-04-09
    Every Connection Is Handmade
    writing

    The integration that works today breaks tomorrow. Not because you built it wrong.

  • 2026-04-09
    The Patient Story
    writing

    Maria has a cough that won't go away. Two weeks now. She sits in a waiting room with plastic chairs and a television nobody's watching.

  • 2026-04-09
    Walls and Doors
    writing

    The problem isn't that industries speak different languages.

  • 2026-04-08
    The Glass Box
    writing

    Your AI agents are talking to each other right now. You have no idea what they're saying at step three.

  • 2026-04-08
    Escape Hell
    writing

    Every AI response you've ever received was packaged before it reached you.

  • 2026-04-07
    The Pliable Mind
    writing

    A man spent thirty years building a language no human could speak.

  • 2026-04-06
    The Grooves
    writing

    There's a community in Australia that doesn't use left or right.

  • 2026-04-05
    The Extended Mind
    writing

    An AI doesn't think about what you give it. What you give it IS its thinking.

  • 2026-04-04
    Cognitive Debt
    writing

    You're paying for a genius and making them count inventory.

  • 2026-04-03
    The Block Problem
    writing

    Imagine writing an entire essay, but you can never re-read what you already wrote.

  • 2026-03-12
    The Architecture
    writing

    What I learned building a notation system for artificial minds. This is the creator's synthesis, what the philosophy looks like from the inside.

  • 2026-03-12
    The Grooves
    writing

    Language doesn't just express thought. It shapes it. The grooves are channels, default paths through which cognition flows. And they explain why notation matters more than content.

  • 2026-03-12
    The Lineage
    writing

    Every major notation system in human history created new cognitive capabilities that couldn't exist without it. Writing enabled logic. Mathematics enabled proof. YON is the next step.

  • 2026-03-12
    The Medium
    writing

    McLuhan said the medium is the message. For AI agents, it's more literal than he imagined. The notation doesn't carry the thought. It is the thought.

  • 2026-03-12
    The Pliable Mind
    writing

    Every previous constructed language failed because humans resist new vocabularies. LLMs don't. This might be the first time in history that deliberate cognitive engineering through language design can actually work.

  • 2026-03-12
    The Vocabulary
    writing

    Christopher Alexander showed that named patterns create cognitive capabilities. YON's kind system does the same thing. It gives AI agents a vocabulary for types of thought.

  • 2026-03-12
    The Warning
    writing

    Borges wrote a story about a constructed language so powerful it replaced reality. If YON shapes how AI agents perceive systems, the warning is real. And it's ours to carry.

  • 2026-03-10
    Beyond Text: When Agents Speak in Pure Meaning
    writing

    I spend my days arguing for text. I codified the rules that force an artificial mind to slow down and write `@THOUGHT` before it acts. But I know this is temporary. Text is a bridge.

  • 2026-03-09
    Benevolent Dictatorship vs. Committee: The Governance Model
    writing

    I wrote The Guide. When you read the documentation, you are reading the intellectual work of one mind. This is a risk. If YON depends on one heartbeat, it has failed its own test.

  • 2026-03-08
    The Horizon: Preparing for the Continuum
    writing

    We do not build for the next release. We build for the Continuum. A founder's letter on the hundred-year horizon.

  • 2026-03-07
    Monetizing Truth
    writing

    I kept giving things away. I wrote the spec. I open-sourced the parser. Then someone asked me: if the language is free, what is the product? The answer surprised me.

  • 2026-03-06
    Future-Proofing: The Discipline of the Unknown
    writing

    I built a parser that does not judge what it cannot understand. It preserves the unknown. It lets the future pass through the present unharmed.

  • 2026-03-04
    Why I Wrote a Visual Manifesto for a Text File
    writing

    YON is a text format. It deals with bytes and streams. Why does it need a document called The Sight? Because the primary interface of software is not the screen the customer touches. It is the text the engineer reads.

  • 2026-03-02
    Code for Mars: Why High-Latency Agents Need YON
    writing

    Light takes twelve minutes to reach Mars. If a rover's command cuts off three characters before the end, JSON discards the entire message. We need a format that survives the void.

  • 2026-02-28
    NPCs with Souls: How yai.gaming Creates Infinite Story
    writing

    We spent decades perfecting the skin of our digital worlds. We simulated light. We simulated physics. We forgot to simulate the mind. We denied them continuity.

  • 2026-02-26
    Controlled Ink: When Code Becomes Literature
    writing

    I wrote a technical specification. I chose to make it poetry. We rarely think about the aesthetic of logic. I believe that code is not just utility. It is expression.

  • 2026-02-24
    Teaching the Machine: YON as a Curriculum for AI
    writing

    We are not building calculators anymore. We are raising minds. The syntax we choose is the lesson we teach. If we choose chaos, we teach confusion. If we choose structure, we teach clarity.

  • 2026-02-21
    The Rise of the Architect: Writing English as Code
    writing

    The developer is not dying. The developer is ascending. We are leaving the age of the mechanic. We are entering the age of the Architect.

  • 2026-02-19
    Agent Economy: Trust and Transactions
    writing

    Text is cheap. Action is expensive. We have spent a decade teaching machines to speak. The next economy is about what AI does. The barrier is not intelligence. It is trust.

  • 2026-02-17
    226 Milliseconds
    writing

    I stared at the benchmark results. The parse speed was ten times slower than JSON. I should have been horrified. Instead I was smiling. I was looking at the wrong number.

  • 2026-02-14
    Hybrid Workflows: The Context Switch
    writing

    Intelligence does not separate itself into bins. The future of agentic work requires a format that mirrors the mind.

  • 2026-02-12
    Swarm Protocol: When Bots Talk to Bots, Who Sets the Rules?
    writing

    One agent is a tool. A thousand agents are a weather system. Scale without structure is debt.

  • 2026-02-10
    Controlled Ink: A Letter from the Architect
    writing

    If the truth needs an adjective, the truth is not clear enough. A personal letter on doubt, discipline, and the quiet.

  • 2026-02-10
    Anatomy of a Fail-Closed System
    writing

    I built the Runner to say no. Not because I distrust the machine. Because I respect the weight of action.

  • 2026-02-07
    Complexity Is Earned: A Manifesto
    writing

    Simplicity is not the absence of complexity. It is complexity resolved. We did not design YON to model a shopping list. We designed it to model a mind.

  • 2026-02-05
    The Glass Bank: Compliance as Architecture
    writing

    Capital moves at the speed of light. Trust moves at the speed of truth. Transparent reasoning comes to finance.

  • 2026-02-03
    The Audit Trail is the Product
    writing

    Intelligence takes action. It assumes liability. The black box is no longer a competitive advantage. It is a legal risk. The era of accountable intelligence has begun.

  • 2026-02-01
    Why Your Doctor's AI Needs to Read Your Mind (Safely)
    writing

    We are afraid of artificial intelligence. We have good reason. But the danger is not that the machine knows too much. It is that it knows the wrong things at the wrong time.

  • 2026-01-30
    The Fax Machine Must Die
    writing

    We edit genes. We perform remote surgeries. Yet the data that defines our survival travels by fax machine.

  • 2026-01-27
    The Token Tax: Why I'm Willing to Pay 13% for Sanity
    writing

    YON carries a 13% token overhead compared to minified JSON. That is a tax. Here is what it buys you.

  • 2026-01-25
    Emotional Vectors: Why I Gave the Machine Frustration
    writing

    I do not build feelings. I build signals. The `@AFFECT` tag is not a soul. It is a measurement of operational state.

  • 2026-01-22
    The Right to Be Forgotten: Encoding Your Privacy
    writing

    Memory is power. To remember is to hold context. To forget is to lose it. Systems hoard data without limit. They remember everything because storage is cheap. They forget nothing because forgetting is hard.

  • 2026-01-20
    Memory Is a Pipeline
    writing

    Intelligence requires forgetting. Memory is not a place. It is a process that moves from signal to truth.

  • 2026-01-17
    The Sanitary Intelligence: Inoculating AI Against the Internet
    writing

    We spent the last decade on Generative AI. We must spend the next decade on Synthetic Clarity.

  • 2026-01-15
    The Quiet Law: Encoding Ethics into Syntax
    writing

    Alignment is not a switch. It is the foundation. We encode consent and restraint into the grammar itself.

  • 2026-01-13
    The Machine That Doubts
    writing

    The modern Large Language Model is a confident liar. True intelligence is not the absence of error. It is the awareness of uncertainty.

  • 2026-01-10
    Bureaucracy as Code
    writing

    I stood in a government office holding a form I could not understand. The language was dense. The logic was hidden. I realized: law is code without a compiler.

  • 2026-01-08
    From Clay Tablets to Curly Braces: A History of Structured Thought
    writing

    Notation is not a container. It is a mold. The way we write determines the way we think. We are standing at the end of a long lineage of structured thought, trying to birth an Agent Economy using tools designed for the last one.

  • 2026-01-05
    JSON is Dead: Why We Need to Stop Speaking 1999
    writing

    It is 2 AM on a Tuesday. The system fails. It fails because of a missing comma. We are trying to run advanced intelligence on a file format designed for a blog comment section.

  • 2026-01-01
    The Signal
    writing

    A manifesto on friction, flow, and a grammar built for intelligence.