Welcome to Dan’s brain

Ergodicity economics

Time averages, ensemble averages, and other ways to disagree about a coin flip

In Which the Divergence Between Ensemble and Time Averages Is Examined Through a Repeated Coin-Flip Wager, Demonstrating How Multiplicative Wealth Dynamics Necessitate the Use of Logarithmic Growth Rates.

Neural generative audio

Wherein Neural Networks Are Taxed With the Generation of Sound, From Raw Waveform to Latent Diffusion, and the Peculiar Matter of a Raspberry Pi Rendering Timbre in Real Time Is Duly Noted.

Generative AI workflows and hacks 2026

Wherein a Running 2026 Log of Generative-Ai Tooling Is Compiled, Cataloguing Local Model Hosting, Provider Training Postures, Markdown-Copy Repairs, and a Fish Script for Salvaging ChatGPT’s Mangled Clipboard Output.

AI search

Retrieval-augmented generation for the working schlub

In Which a Personal Repository Is Queried Through Both Classic Term-Frequency Statistics and Modern Neural Embeddings, Demonstrating That Local Search Efficacy Is Achieved Without a Dedicated Database.

Code agents and assistants

Turing-complete autocorrect, vibe-coding, …

Wherein Various Coding Harnesses—cloud, Offline, and Terminal-Based—are Surveyed and Compared, With Particular Attention Paid to Whether Each Agent Relies Upon Grep, Embeddings, or Repo-Maps to Locate Relevant Code.

Taking notes

Notes on notes on notes

In Which the Art of Note-Taking Is Dissected Through Various Philosophies and Digital Architectures, With Specific Attention Directed Toward the Inherent Tension Between Offline Data Security and Mobile Utility.

Configuring my code with “.env” files

Wherein the Author Abandons Python’s Dotenv for the Shell-Native Tool Direnv, and Details How Per-Directory .Envrc Files Silently Load and Unload Variables Upon Entering and Leaving Folders.

Voice transcriptions and speech recognition

Wherein Dictation Tools, Voice-Driven Coding Platforms, and Whisper-Based Transcription Services Are Surveyed, With Particular Attention Paid to the Vexing Matter of Software That Silences One’s Background Music During Use.

Universal artificial intelligence, AIXI

Wherein AIXI Is Presented as an Uncomputable Yardstick of Perfect Rationality, Combining Solomonoff Induction With Reward-Maximizing Policy, and Wherein Schmidhuber’s Speed Prior Is Offered as a Computable, Though Hardly Cheap, Cousin Thereof.

Web search for machines

Search and scraping APIs for agents, which do not click links

Wherein It Is Considered How Machines, Unlike Men, Decline to Click Links, and How Vendors Such as Jina, Firecrawl, and Tavily Do Compete to Furnish Them Pre-Chewed Markdown Instead.

Internet search engines

Tips, tricks, confidentiality

Wherein the Decline of Commercial Search Engines Is Chronicled, and Remedy Is Sought Among Privacy-Minded Alternatives, Metasearch Proxies, and AI-augmented Upstarts, With Kagi and Marginalia Noted as Curious Survivors.

Fish shell

A command line shell that does not think that I am the problem

Wherein the Fish Shell Is Examined as an Alternative to Bash, Its Opinionated Design, Sarcastic Documentation, and Practical Configuration of PATH Variables, Plugins, and Python Environments Being Treated at Length.

Optionality as an end in itself

On optimizing for not optimizing

Wherein the Moral Status of Preserving Future Possibility-Space Is Examined Through Three Formal Frameworks, Including Empowerment Theory, Ergodicity Economics, and Quality-Diversity Algorithms.

PDF, Portable Document Format

How we may use a thousand dollar computer to simulate a one cent piece of paper with zero day exploits

Wherein the Peculiarities of the Portable Document Format Are Examined, With Particular Attention to Its Manipulation by Command-Line Tools, Python Libraries, and Other Instruments of Varying Reliability

Practical cloud machine learning

Cloudimificating my artificial data learning intelligence brain clever science analyticserisation

Wherein Two Approaches to Cloud GPU Access Are Distinguished, the Rented Remote Machine and the Serverless Function, and a Hybrid Hyperparameter-Optimisation Workflow Is Described.

Generative art with language+diffusion models

also some autoregressive models

In Which the Prompt-to-Image Model Ecosystem Is Surveyed and the Legal Frameworks Governing AI Image Training and Publication Across Japan, Singapore, the EU, and Australia Are Examined.

Fine tuning foundation models

Wherein Language Models Are Aligned With Human Intent Through Reinforcement Learning and Low-Rank Adaptation, and Wherein Refusal Is Found to Be Mediated by a Single Direction in the Residual Stream.

Mathematics without LaTeX

Wherein the Rendering of Mathematical Notation in Web Documents Is Surveyed, With Particular Attention Paid to Server-Side Pre-Rendering as an Alternative to Client-Side JavaScript Libraries.

Zotero

The adequate citation tool

Wherein a Browser Plugin Is Employed to Capture Article Metadata and PDFs in a Single Click, BetterBibTeX Citation Keys Are Generated, and Local REST APIs Are Described.

Care and feeding of macOS filesystems

Wherein the Complexities of Filesystem Compatibility Between macOS and Foreign Systems Are Examined, With Particular Attention to Unicode Normalisation Hazards Encountered During Cross-Platform File Transfer.

Virtual private networks

More options for internet privacy

Wherein the Distinction Between Anonymous Access and Private Mesh Networking Is Drawn, the DNS Leakage Attendant Upon OpenVPN Under Linux Is Described, and Methods of Split Tunneling by Platform Are Surveyed.

Markov decision problems

Wherein a Beer Warehouse Inventory Problem Is Traced From a Simple MDP to a Multi-Agent Decentralised POMDP, With Belief States and Bellman Recursion Introduced Along the Way

Editing images with AI

Wherein a Distinction Is Drawn Between Instruction-Following Editors and Single-Task ML Tools, With Attention to Specialised One-Trick Startups Now Rendered Redundant by General-Purpose Models.

Auditory features

descriptors, maps, representations for audio

Wherein the Extraction of Compact Audio Descriptors Is Catalogued, and It Is Noted That Self-Supervised Networks Such as HuBERT Now Furnish Discretized Codebooks Read Directly by Audio Language Models.

Neural codecs and compression algorithms

Neural bandwidth reduction

Wherein Neural Networks Are Enlisted to Compress Images and Audio, Notably an Audio Codec Folding Frozen HuBERT or WavLM Features Into Its Residual Codebooks So Tokens Carry Semantic, Not Merely Spectral, Meaning.

Voice fakes

Wherein the Reader Is Guided Through the Long, Tangled History of Voice Synthesis and Cloning, From an 1860 Talking Machine to Modern Neural Vocoders, With Due Skepticism Toward Proprietary Demos.

The language game

Coevolution of words and meanings

Wherein Semantics Is Recast as a Classifier Problem, With Word-Meaning Defined via Divergence Between Two Speakers’ Corpora, and the Vexed Status of Cheetahs as Cats Duly Considered.

Multi-objective optimization

Wherein the Reader Is Warned That Summing Weighted Objectives Into One Score Discards Pareto-Optimal Points on Concave Stretches of the Front, However Cleverly the Weights Are Tuned.

Nostr

Social media for the stroppy

Wherein a Minimalist Relay-Based Protocol Is Examined, Its Modest User Counts Noted, and Its Fragmented Approach to Group Messaging Considered Against the Promise of Censorship Resistance.

Goodhart’s Law

also Campbell’s Law, and the Overfit Theory of Everything

Wherein the Failure of Proxy Measures Is Examined, With Attention to Four Distinct Mechanisms by Which Optimisation of a Target Is Found to Corrupt Its Original Purpose.

AI usage policy

I have already given up on my one-person crusade to have this called a “slopicy” but you can pleasantly surprise me

Wherein the Author’s Varied Applications of Generative AI to Blog Maintenance, Literature Review, and Self-Tutoring Are Enumerated, With Notes on Labelling Practices for Robot-Assisted Content.

Semantics

Compressed representations of reality for syntactic agents; which might be what meaning means

Wherein the Mapping of Words to Phenomena Is Treated as a Classification Problem, and Shared Human Experience Is Proposed as the Mechanism by Which Mutual Understanding Between Speakers Is Achieved.

Sufficient statistics

Summarizing the data with a minimal representation

Wherein the Canonical Sufficient Statistic Is Revealed to Be a Probability Measure, and Its Many Aliases Across Statistics, Control Theory, and Machine Learning Are Catalogued.

AI Agents for science

Outsourcing knowledge of base reality to bots

Wherein the Scientific Literature Is Navigated by Autonomous Agents, With Particular Attention to Systems That Read Thousands of Papers and Synthesise Citations Without Human Intervention.

Sandboxing apps

Upon letting strangers into your house, just so long as they don’t open the door to that one forbidden room

Wherein the Isolation of Untrusted Applications Is Examined, With Attention to the Practical Weaknesses of Flatpak, Firejail, Bubblewrap, AppArmor, and macOS Seatbelt.

Operationalising the bitter lessons in compute and cleverness

Amortizing the cost of being smart, applied resource-rationality

Wherein the Bitter Lesson Is Reframed as a Substitution Curve Between Compute and Cleverness, Its Slope Being Found to Vary Across Data-Rich and Data-Scarce Regimes Rather Than Holding Universally.

Groupthink and the wisdom of crowds

God help me I need to extract truth from the internet

Wherein the Conditions Under Which Diversity Among Group Members Is Found to Yield Measurable Epistemic Dividends Are Examined, Alongside Mechanisms Such as the Surprisingly Popular Algorithm.

Continual learning in neural nets

Also catastrophic forgetting, catastrophic interference, lifelong learning, …

Wherein the Challenge of Catastrophic Forgetting in Sequential Task Learning Is Examined, and Regularization, Replay, and Dynamic Architecture Approaches Are Surveyed Alongside a Bayesian Framing of the Problem.

Neural denoising diffusion models of language

Wherein Bidirectional Attention Is Shown to Enable Native Text Infilling, and DiffusionGemma Is Presented as a Concrete Instantiation of Discrete Diffusion Over a Jointly Denoised Token Canvas.

Performative prediction

A thing people worry about when they worry about “socially constructed” categories

Wherein a Credit Classifier Is Found to Cause the Defaults It Predicts, Performative Stability Is Distinguished From Optimality, and Strategic Manipulation of Features by Applicants Is Examined.

Vibecoding Apple apps

Shipping macos/iOS automation without learning Swift

Wherein Two MCP Servers Are Set to Work in Concert — One for the Build Engine, One for the Running IDE — So That Personal iOS Applications May Be Constructed Without Swift Being Learnt

Backups

Version control for horrible data

Wherein the Relative Merits of Cloud and Local Backup Solutions Are Surveyed, With Particular Attention Paid to the Chronic Unreliability of Bus-Powered Spinning Drives.

TikZ/PGFplots etx

Wherein TikZ Is Compiled to WebAssembly and Rendered in the Browser Without a LaTeX Installation, Its Portability Across Quarto, Obsidian, and Other Environments Is Examined, and Various GUI Frontends Are Noted.

Institutions for angels

How open movements get captured, hijacked and hollowed out (which is totally OK sometimes)

Wherein Open Membership Is Examined as a Structural Vulnerability, and Adverse Selection, Entryism, Elite Capture, and the Tyranny of Structurelessness Are Catalogued as Its Predictable Consequences.

Game theory

Wherein Equilibrium Concepts Are Examined Across Cooperative and Non-Cooperative Branches, the PPAD-completeness of Nash Equilibrium Computation Is Remarked Upon, and Behavioural Extensions Are Surveyed.

Marimo

A python visual notebook that works like I imagined scientific notebooks should

Wherein a Python Notebook Environment Is Examined, Its Notebooks Being Stored as Decorated Plain Python Files Rather Than JSON, and Cell Execution Order Being Determined by an Inferred Dependency Graph.

DNS

On asking strangers for directions

Wherein the Privacy Implications of DNS Resolver Selection Are Examined, and the Installation of Encrypted-Transport Profiles Upon Apple Devices Is Described as a Remedy for ISP Observation of Browsing Habits.

Editing images using code

Painting with a hammer, blindfolded

Wherein Several Image-Processing Tools Are Surveyed, the Rivalry Between ImageMagick and GraphicsMagick Is Remarked Upon, and Vips’s Demand-Driven Pipeline Approach Is Described.

Adaptive design of experiments

I am not going to call it ‘Bayesian optimization’, but that is what everyone else does

Wherein the Sequential Optimisation of Costly Black-Box Functions Is Considered, Acquisition Functions Are Introduced as Guides to Sampling, and Gaussian Process Surrogates Are Employed Toward This End.

Grass-roots friendly societies could provide cooperative insurance against dark times

Let us run ahead of the future and start building the institutions we want to have

Wherein a Small Melbourne Society Is Proposed, Funded at $200 a Month, Its Pooled Fees Are Invested in Countercyclical ETFs, and Compliance Costs Are Reduced by AI to $41k–56k Yearly.

Taking my email archive offline

Wherein a 20GB Correspondence Trove Is Drawn Down by Mbsync Into an Encrypted Maildir, Is Replicated by Syncthing, and the Server Is Left With a 30–90‑day Window for Phone Use.

Science Fiction

increasingly, now-casting

Wherein a Private Canon of Science Fiction Is Sketched, Recommendations Being Filtered by Audiobook Availability, and a Prefigurative AI-apocalypse Movie Night Being Arranged.

New project: an Alignment Journal

Wherein an Open-Access Alignment Journal Is Founded, Alternate Peer Review Is Trialed, and a One-Year Pilot Is Conducted by Two Managing Editors, and Submissions Are Sought Through a Mailing List.

Git tricks

Wherein Sundry Git Particulars Are Gathered, and a Merge’s Disputed Files Are Set Wholesale to Ours or Theirs by Diff Piped to Checkout, With Fish Aliases Kept for Later Remembrance.

Evolution strategies

Slimmest possible implementation of “evolutionary” optimization

Wherein Gaussian Perturbations Are Applied to Neural Weights, Fitnesses From Many Forward Passes Are Gathered, and an Antithetic-Pair Update Is Formed Without Any Backward Computation.

Running LLMs locally on a Mac

Chasing the visceral sense of LLM effort that comes from your laptop ironing out your trouser crease

In Which the Local LLM Ecosystem on Modern Apple Silicon Is Dissected, Revealing a Landscape of Specialized Runtimes, Mixed-Precision Weight Formats, and the Specific Memory-Management Taxes of Local Inference.

Building a maths agent

Delegating the queen of the sciences to the machines

In Which an Iterative Harness for Mathematical Reasoning Is Constructed, Wherein the Significance of a Robust Execution Loop and Scratchpad Memory Is Weighed Against the Simple Cost of Model Fan-Out.

Common knowledge/shared knowledge

Wherein the Distinction Betwixt Shared and Common Knowledge Is Illustrated by Voting Polls and the Weinstein Disclosures, Before a Bayesian Formalisation Involving Σ-Algebras and Hopping Agents Is Offered, Unbidden.

Incoming links and notes

Wherein the Author Amasses an Ever-Growing Ledger of External Links, Bookmarks, and Marginal Notes, Spanning AI Safety, Rock Illustrations, and Whale Linguistics, Left Deliberately Unsorted for Future Attention.

Probably actually reading/writing

Wherein the Author Catalogues a Sprawling Personal Notebook of Unfinished Inquiries, Spanning Bounded Agency, Coalitional Fairness, and House Renovation, and Confesses That Many Fields Are Already Mined-Out.

Agent harness design

The weird technologies that make language models do things

Wherein the Harness Surrounding a Language Model Is Dissected Into Wiring, Verification, Context, and Dispatch, With the Compiler-Checked Lean Theorem Prover Offered as the Cleanest Case Study of the Whole Arrangement.

Fine-tuning danbot

Wherein the Author Trains a Small Language Model on Paired Samples of His Own Prose Against AI-generated Slop, Finding a Stubborn Detector-Versus-Taste Gap That Persists at Forty Held-Out Paragraphs.

Agent frontends

Where the human sits, and what the chat window has to be able to do

Wherein the Merits of Chat Frontends Are Weighed by Whether They Expose Sampling Parameters, Revealing That Open WebUI’s Model Presets Alone Justify Its Notoriously Fussy Node Dependencies.

Reasoning and proof models

Inference-time-scaling considered unreasonably accurate

Wherein Is Charted the Design Space of Mathematical Reasoning Systems—solvers, Provers, and Autoformalizers—noting That a Lean Compiler’s Exact Verdict Permits Sampling Budgets Into the Thousands, Unlike Majority-Vote Solvers.

AI agents, applied

Stack vocabulary, MCP, and the products that wrap the models

Wherein the Reader Is Furnished a Shopping List of Agentic Harnesses, Wherein the Trade-Offs Between Observability, Blast Radius, and Attention Budget Are Weighed, and MCP and Skills Protocols Are Duly Distinguished Before Recommendations Are Ventured.

Czechia

In Which the History of Czechia Is Surveyed Through Its Notable Exports, From Hussite War-Wagons and Defenestrated Diplomats to Pilsner Lager and the Word Robot.

Depression

Wherein the Challenges of Supporting Depressed Loved Ones Are Examined, With Particular Attention Paid to Circular Reasoning Patterns and the Bounds of Non-Professional Assistance.

Open ended intelligence

Algorithms worth running for a billion years

Wherein the Concept of Open-Endedness Is Examined as an Alternative to Single-Point Optimisation, With Reference to Algorithms Designed to Generate Novelty Indefinitely Rather Than Converge Upon Fixed Goals.

The unreasonable effectiveness of empirical risk minimization

General methods that leverage computation with scalar loss functions and a static data distribution are ultimately the most profitable, and by a large margin

Wherein the Tension Between the Tractability of Empirical Risk Minimisation and the Open-Ended, Path-Dependent Nature of Human Life Is Examined, With Scalar Loss Functions Found to Impose a Static Distribution Upon an Irreducibly Contingent World.

Wisdom from the madness of crowds

Social belief with cheapo Bayes-alikes

Wherein the Pragmatist Notion of Beliefs as Decision-Driving Mechanisms Is Extended to Bounded Social Agents, Who Are Seen to Import the Beliefs of Others Rather Than Derive All Knowledge Independently From Evidence.

Cooperation through uncertainty

Distributed sensing, swarm sensing, adaptive social learning, multi-agent adaptation, iterated game theory with learning etc. tl;dr: ignorance is strength

Wherein the Signal-Detection Problem of Kin Recognition Is Examined, and an Acceptance Threshold Is Derived by Which Noisy Cues of Relatedness Are Weighed Against Fitness Costs and Benefits

Models of inequity

Sharpening the fuzzy process of bias, discrimination and inequity

Wherein Agent-Based and Game-Theoretic Models of Cumulative Disadvantage Are Collected, and the Glass Ceiling Effect Is Shown to Arise From the Compounding of Individually Minor Discriminatory Mechanisms.

Bikes

Especially bikes where I live, which means Melbourne at the moment

In Which a Cost Comparison of Urban Cycling Against Car Ownership Is Conducted, a Net Annual Saving of Some AUD9,500 Being Calculated When Travel Time Is Counted as Cardio Exercise.

Metis and .*-rationality

High modernism, spontaneous order, legibility, the Great Society, technocracy, local knowledge

In Which High Modernism’s Erasure of Tacit Local Knowledge Is Examined, and Bullet-Proof Ritual Powder Is Offered as Evidence That Genuine Rationality Is Concealed Within Apparently Irrational Belief.

Quarto

Swiss army knife for hand-whittling scientific reports, slides and blogs

Wherein a Pandoc-Based Scientific Publishing System Is Examined as a Unifier of Jupyter Notebooks and R Markdown Workflows, Its Support for Multiple Computational Languages and Diverse Output Formats Being Noted.

Decision theory in mechanized causal graphs

Newcomb’s box via the back door, the intentional stance for arrow diagrams

Wherein Mechanism Nodes Are Introduced as First-Class Elements of Causal Graphs, and Newcomb’s Problem Is Examined Such That a Predictor Is Seen to Read an Agent’s Policy Rather Than Its Act.

Chromium browsers

In Which the Chromium Browser Engine Is Surveyed, With Particular Attention Given to the Privacy Shortcomings of Google Chrome and Several Alternative Browsers Examined as Potential Remedies.

Why does deep learning work?

Are we in the pocket of Big VRAM?

In Which Several Competing Explanations for the Effectiveness of Deep Learning Are Surveyed, Including Connections Drawn Between Stochastic Gradient Descent and Statistical Mechanics.

Effective altruism

First world trolley problems

Wherein the Opportunity Costs of Charitable Giving Are Examined, and the Tension Between Low-Variance Incremental Donations and Hits-Based Portfolio Approaches Is Surveyed.

Hollow states

Skim government when fat cats drank the cream

Wherein a Citizen’s Return on Investment in a State Captured by Kleptocrats Is Set Against the Cost of Its Overthrow, and Modern Mechanisms of Capture Are Treated.

Markdown editors and viewers

Wherein the Requirement for Mathematical Equation Rendering Is Established as a Criterion for Editor Selection, Note-Taking Applications Are Found to Serve Double Duty, and Various Tools Are Enumerated.

Snowmobile or bicycle?

Complement or substitute?

Wherein Krakauer’s Distinction Between Complementary and Competitive Cognitive Artifacts Is Applied to Modern AI, and the Skolt Saami Snowmobile Case Is Invoked as a Precedent for Collective Atrophy.

Organising a photo collection

Wherein Cataloguing Software Across Several Platforms Is Surveyed, and Tools for Stripping Embedded Metadata — GPS Coordinates, Timestamps, and Camera Serials — From Photographs Before Sharing Are Examined.

Rust

The programming language that the cool kids are rewriting everything in

Wherein the Incursion of a Compiled Systems Language Into Python Tooling Is Surveyed, and Its Extension Into Browsers, Editors, CLI Utilities, and Desktop Environments Is Further Noted.

AI disempowerment of humans

Races to the bottom in human relevance, gradual disempowerment

Wherein the Process of Gradual Disempowerment — Driven by Feedback Loops and Competitive Selection Rather Than Sudden Catastrophe — Is Traced Across Economic, Epistemic, and Political Domains.

Hosted Functions / Serverless

Wherein Functions Are Submitted to a Platform, Invoked Upon Triggering, and Charged Per Execution, With Idle Server Costs Being Traded for the Occasional Latency of Cold Starts.

DIY social networks and social-groupware

V. cypherpunk! much 733+! Underground!

Wherein Options Ranging From Commercially Hosted Chat Services to Fully Serverless Peer-to-Peer Networks Are Catalogued, and the Question of Participant Critical Mass Is Noted.

Economics of cognitive and labour automation

Practicalities of competition between ordinary schlubs, and machines which can tirelessly collage all of history’s greatest geniuses at once

Wherein the Automation of Cognitive Labour by Large Language Models Is Surveyed, and a Consequent Shift in Organizational Incentives From Open Knowledge Toward the Private Sequestration of Resources Is Noted.

Docker containerized apps (for scientists)

Doing things that previously took 0.5 computers using 0.4 computers

Wherein Docker’s Terms Are Disentangled, Dockerfiles Are Prescribed for Repeatable Environs, and an Opaque Registry Timeout Is Remedied by Setting the Daemon’s DNS to 8.8.8.8.

Unix/linux distros explained as bikes

Wherein a Bewildering Family Tree of Unix Breeds Is Rendered as Bicycles; Wheels Are Withheld for Freedom, Parts or Files Are Dispatched for Assembly, and Models Are Replaced on Schedule.

Terminals

More realistically, terminal emulators, because anyone who owns an actual working VT102 these days is not using it to do work

Wherein Terminal Emulators Are Surveyed, Graphics Being Rendered by Sixels or Chafa, and Hyper’s URL Links Being Restrained by a Ctrl‑Click Setting, Amid Sundry Multiplexed Contrivances.

How to reduce government spying on me

Wherein the Citizen’s Traffic Is Obscured by Hidden DNS Queries, Bluetooth Is Kept Silent, and Papers Are Conveyed by SecureDrop’s Onion Gate, Whilst Civic Leagues Are Supported.

Time management

TO DO lists, task scheduling, and the psychology and practicality of getting stuff done

Wherein Interruptions Are Shunned, Devices Are Timed by ActivityWatch, and Procrastination Is Restrained by Beeminder-Style Commitment Contracts, With Losses Being Pledged to Keep Work in Motion.

ssh

a.k.a. “Secure shell”, which is not actually a shell but an encrypted network protocol that people like to shell with

Wherein Sundry Secure Conveyances Are Effected by Client Contrivance, the EDITOR Is Set by SSH_CONNECTION, and the Session Is Suspended by Enter ~ and Ctrl‑Z.

Intentional language

The point of teleology, Cognitive ergonomics of anthropomorphising

Wherein the Wason Card Puzzle Is Recast as Underage Drinking, and Dennett’s Intentional Stance Is Employed to Ease Prediction of Machines, While Moral Patienthood and Pareidolic Overshoot Are Set Aside.

PIBBS x ILIAD Research residency January-February 2026

Wherein a Shoreditch Residency at the London Initiative for Safe AI Is Recounted, Held 5 January–13 February 2026, With Lectures Ranging From Information Decomposition to Singular Learning Theory.

Disseminating science

Journals and preprint servers etc

Wherein the Economics of Journals Are Surveyed, Impact Indices Are Recited, and the Resort to Shadow Libraries Such as Sci‑Hub and Anna’s Archive Is Noted in the Scholar’s Daily Traffic.

Money, Australian-style

Wherein the Colonist’s Accounts Are Kept, Capital Gains Are Traced From Cryptic Share Codes, AUSTRAC-registered Coin Exchanges Are Named, and BAS Ledgers for the Sole Trader Are Ordered.

How to communicate

with humans

Wherein Assertive, Non-Violent, and Decoupled Talk Is Catalogued; Ask-Versus-Guess Norms Are Contrasted, and Crunch-Mode Costs Are Estimated as Leave and Recovery After Prolonged Sprints.

Sleep

Pathologies and hacks

Wherein the Uncertainties of Sleep Science Are Surveyed, Blue Light and Melatonin Timings Are Noted, and Jet Lag Is Managed by Timeshifter’s Commands of Sun, Shade, Coffee, and Naps.

Attention economy

Wherein Attention Is Framed as a Rivalrous Scarce Resource, Is Formalized as a Constraint in Optimization, Are Platform Incentives Described to Capture User Time, and Are Billboards’ Links to Road Fatalities Noted.

Bayes inference in an open world

Realizability, infrabayesianism, M-open, M-closed, mis-specification

Wherein Misspecified Bayes Is Treated as M-Open, and Predictive Mixtures Are Formed by LOO Cross-Validation With PSIS, While Likelihoods Are Tempered by Η to Restrain Overconfidence.

Hierarchical reinforcement learning

Skill discovery, abstraction discovery, concept induction for active agents

Wherein the Problem of Long Horizons Is Addressed by Decomposing Tasks, and Internal RL Is Introduced Whereby a Meta‑controller Is Employed to Manipulate Model Residuals Sparsely, Compressing Token Horizons.

Probability, Cox-style

Wherein an Epistemic Foundation of Probability Is Presented, Wherein Cox’s Desiderata Are Shown to Yield the Product and Sum Rules and Bayes’ Rule Is Shown to Follow, Contrasted With Kolmogorov’s Measure Approach.

Top influences of 2025

Content that changed my life this year, and which also might change yours

Wherein the Author’s Year’s Influences Are Catalogued, and It Is Noted That AI Tools Are Used More Than Ever to Assist Reading, With Consequential Reorientation Toward the Study of Human–AI Interfaces Being Effected.

Stochastic parrotology

Telegraph octopuses, observational world models, conduit metaphors and the limits of reasoning from pre-training

Wherein the Question of Whether Pre‑training Yields Causal Inference Is Examined, and the Possibility That Models Extract Intervention Distributions From Text Corpora and Form Simulators via a Kolmogorov‑like Structure Function Is Considered.

Garbled highlights from Neurips 2025

Wherein Attendance at NeurIPS in San Diego Is Recorded, US and Chinese Paper Counts Are Noted to Be Nearly Equal, a Post‑AGI Workshop on Economics Is Summarised, and a Bespoke Semantic Paper‑search Tool Is Described.

Agency under bounded compute and information

Between the Good Regulator and the Big World

Wherein a Taxonomy of Eight Rival Frameworks for Bounded Agency Is Assembled Into a Table, and It Is Argued That Unlimited-Compute Demons Require, Absurdly, a Larger Universe to Think In.

Economics of foundation models

Microeconomic of compute

Wherein the Reader Is Guided Through a Hub of Sundry Economic Matters Attending Foundation Models, Wherein One George Hosu Is Quoted Remarking Upon the Peculiar Marketing Allure of Chatbots Over Decades of Unadopted Medical Software.

How to reduce criminals spying on you

Wherein the Reader Is Cautioned Against Password Recycling and Phishing Schemes, With a Notable Aside Upon Evilginx2, a Tool by Which Two-Factor Authentication Is Rendered Insufficient.

Community

Engineering, maintaining, organizing, engineering oxytocin and dopamine…

Wherein the Practical Methods of Community-Building Are Surveyed, With Attention Given to the Concept of Asabiyyah and to Resources Offered by Hobby Groups, Organizers, and Consultancies Such as People and Company.

Bayesian elicitation

Information elicitation, incentive mechanisms for truth, proper scoring rules…

Wherein the Design of Incentive Mechanisms for Truthful Reporting Is Examined, With Particular Attention Being Paid to Proper Scoring Rules and Their Connection to Machine-Learning Loss Functions.

Who I donate to

Expensive virtue signalling

Wherein a Preference for High-Variance, Systems-Changing Donations Is Described, With Recipients Spanning Housing Advocacy, Civil Liberties, Open-Source Software, and the Arts.

Delegation and coordination using AI

I’ll give utopia when you take it from my cold, dead, invisible hands

Wherein the Coasean Singularity Is Examined as a Possible Consequence of Deploying AI Agents as Personal Fiduciaries, and the Merging of Economic Negotiation With Democratic Consensus-Building Is Considered.

Renormalization for ML

A Bayesian's dictionary for renormalization-group vocabulary

Wherein a Vocabulary Mapping Between Renormalization-Group Physics and Bayesian Machine Learning Is Presented, With Depth Identified as the RG Flow Itself and Universality Classes as Basins of Fixed Points.

Intrinsic motivation

Do agents learn to want freedom? Do they learn to want to learn? Do they learn to goof off? etc

Wherein Several Formalisations of Self-Generated Reward Are Surveyed, Including Compression Progress, Empowerment, and Novelty Search, Each Resolving to a Choice of Mutual Information Term to Optimise.

Awesome illustrators and printmakers

Raiding history for clip art

Wherein a Personal Canon of Engravers and Caricaturists Is Surveyed, From Dürer’s Armoured Rhinoceros to Gillray’s Tiny Bonaparte, With Links to Rijksmuseum and Other Public Collections.

PDF ingestion

Feeding technical documents to local language models

Wherein the Conversion of PDF Documents to Text Is Examined, and Whole-Page Vision Models Are Found to Corrupt Mathematical Notation Silently, With Specialist Pipeline Converters Offered in Contrast.

Front-end clients for AI image models

ComfyUI, InvokeAI, Draw Things, ChaiNNer and friends

In Which Several Front-End Image-Generation Clients Are Compared on Apple Silicon, and the Choice Between PyTorch, MLX, and CoreML Runtimes Is Found to Hinge on Available RAM and Whether LoRA Adapters Are Required.

London

Wherein the City of London’s Millennium-Old Corporate Privileges Are Examined, Canal-Boat Housing Pressures Are Observed, and Supermarket Choices Are Found to Encode Social Class.

Multi-Agent Reinforcement Learning

Distributed sensing, swarm sensing, adaptive social learning, multi-agent adaptation, iterated game theory with learning etc

Wherein the Formation of Agent Coalitions Is Treated as a Learning Problem, Classical Exponential Complexity Is Noted, and Cooperative Game Theory Concepts Including the Nucleolus Are Adopted.

Australian authoritarianism

Beneath the beach, the barbed wire

In Which the Tension Between Australia’s Self-Image as a Nation of Larrikins and Its Obedience to Impersonal Authority Is Examined, With Attention Paid to Surveillance Legislation and the Prosecution of Whistleblowers.

Cosmic decision theories

Newcomb’s boxes etc

In Which the Distinction Between Causal and Evidential Decision Rules Is Examined Through the Lens of Mutually Predictive Agents, With Newcomb’s Paradox Presented as the Animating Illustrative Case.

An orderly retreat from economic relevance

Red-teaming short-term human purpose

Wherein Several Conditions Under Which Human Labour Retains Competitive Advantage Are Surveyed, Embodied Physical Presence and Freedom From Scarce-Mineral Dependency Being Chief Among Them.

Flashcards

Spaced repetition systems, esp Anki

Wherein Personal Notes on Spaced Repetition Tools Are Presented, Anki Is Treated as the Central Application, and Its Connection to AI Assistants via MCP Servers Is Among the Topics Covered.

Here’s how I would do art with machine learning if I had to

Wherein the Pre-Diffusion Lineage of Generative Neural Art—DeepDream, GANs, CPPNs, and Sketch-Rnn—is Surveyed as a Historical Record, Distinct From Modern Text-to-Image Systems.

Aligning AI systems

Practical approaches to domesticating wild models. RLHF, Constitutional AI, etc

Wherein the Notion of Alignment Is Considered Systemic, the Human Corpus, Pretraining, RLHF and Deployment Are Treated as a Single Epistemic Ecosystem, and Institutional and Economic Dimensions Are Noted.

Reputation systems

Karma, credit scores, PageRank, proof-of-identity, proof-of-truth; optimized ad hominem reasoning

Wherein the Measurement of Trustworthiness Through Iterative Peer Endorsement Is Examined, Such That One’s Rating of Another Is Weighted by the Ratings Accorded to Oneself.

Multi-agent RL tooling

Simulation environments and frameworks for multi-agent research

Wherein Various Platforms Are Surveyed, Including Open-World Game Engines and Social Dilemma Suites in Which Agents Are Evaluated on Their Capacity to Cooperate With Previously Unseen Partners.

The public sphere and its business models

Free speech versus market-clearing speech

Wherein Ostrom’s Framework for Sustainable Commons Management Is Applied to the Problem of Online Discourse, and the Tension Between Advertising-Funded Platforms and Judicious Public Debate Is Examined.

Virtual private mesh networks

Pretending your phone is on your LAN

Wherein Secure Private Meshes Are Surveyed for Use in Secure Access Rather Than Anonymity, and It Is Noted That Tunnel Brokering Is Performed by Providers Who Observe Which Devices Interconnect, Authentication Being Routed via Corporate Identity Providers.

Categorical systems theory

String diagrams, maybe other useful things

Wherein Systems’ Interfaces Are Treated as the Sole Locus of Interaction, Composition Patterns Being Prescribed; String Diagrams Are Employed for Bayesian-Network Computations, With DisCoPy and TikZiT Noted.

Blame engineering

On institutions for siphoning responsibility

Wherein Culpability Is Apportioned Between Persons and Institutions, Accountability Sinks Are Surveyed, and Blame Is Shown to Be Diffused by Bureaucracy, From Aviation Post‑mortems to Borderless Porn Regulation.

How to reduce corporate spying

on me, hopefully

Wherein Ambient Data Leakage Is Reduced by Permissions, Browsers, and Networks, and Wherein macOS App Launches Are Noted by Unencrypted OCSP Hashes Sent via Akamai, With Purchases Dispersed Beyond Amazon.

Evolution strategies

Wherein a Neural Net Is Trained Without Backprop by Gaussian Perturbations, Fitness Differences, and Antithetic Pairs, the Same Minibatch Being Shared Across a Population to Temper Variance.

Embedded agency

What about agents that live in the world?

Wherein the Doctrine of Infinite‑compute Agents Is Surveyed, AIXI‑like Worlds Being Shown to Spawn Löbian Self‑reference as Agents Simulate One Another, and the Matter Is Set Aside for Want of Finite Budgets.

Travel hacks

Wherein the Traveller’s Burdens Are Lightened by Ranger-Rolled Shirts, Local SIMs Being Preferred to Costly Travel Ones, and the Head Being Lashed Upright to the Seat by Mask and Harness.

Let’s solve social event organising

Wherein the Problem of Arranging Casual Gatherings Is Considered, and Calendar-Link Generation and Niche Invitation Services Are Offered as Means to Coordinate Dates and Collect RSVPs.

Secure chat systems

Optimizing back channel interjections into other people’s meetings

Wherein Signal Is Recommended for Practical Use, Its Phone‑number Registration Is Noted, Push Notifications Are Mentioned as a Potential Leak, and Face‑to‑face Conversation in a Faraday Cage Is Advised.

Scientific writing

In which tips are given for the projection of status through nominal phrases and passive voice

Wherein the Rituals of Academese Are Examined and the Tradeoff Between Clarity and Status Signalling Is Described, With Attention Paid to Mathematical Notation and Editorial Incentives.

AI evals

Psychometrics for robots

Wherein the Distinction Between Benchmarks and Evals Is Laid Out, and the Rise of Meta‑evaluations Such as EvalEval and Tools Like Inspect Are Noted, With Human Judgements and User Logs Being Employed.

Reinforcement learning

Wherein Donald Michie’s Matchbox Machine, MENACE, Is Described as Teaching Tic‑tac‑toe by Adjusting Bead Counts, and Policy Gradients, Q‑learning and Exploration Schemes Are Laid Out in Formal Terms.

Imprecise Bayesianism

Wherein Imprecise Bayesianism Is Presented as an Alternative for M‑open Problems, Where Beliefs Are Represented by Convex Sets of Distributions and PAC‑Bayes Generalization Bounds Are Invoked.

Multi agent causality

Game theory and decision theory for lots of interacting agents

Wherein Causal DAGs Are Extended to Include Agents and Decisions via a Mechanized Multi‑Agent Influence Diagram, and Iterated Games Are Employed to Exemplify Commitment Races Relevant to AI Safety.

Causal inference in highly parameterized ML

Wherein Causal Graphs Are Applied to Nonparametric Neural Nets, Practical Tooling Such as DoWhy and TETRAD Is Noted for Supporting Explicit Causal Modelling, and Benchmarks Such as CauseMe Are Cited for Dataset-Shift Challenges.

Visualising probabilistic graphical models

Also related models, such as Neural nets

Wherein Probabilistic Graphical Models Are Surveyed and Practical Tooling for Directed Flow Graphs Is Catalogued, With Particular Attention Given to Plate Notation, Inline Math and SVG/PDF Export.

Sciences of the Artificial

What is science for contingent and constructed things?

Wherein Simon’s Argument That Designed Systems Are Best Studied Empirically Is Presented, and Nearly‑decomposable Hierarchies Are Offered as a Means to Model Complexity in Organizations.

Flavours of Bayesian conditioning

Conditional expectation and probability

Wherein Jeffrey Conditioning Is Presented as Rescaling Partition Weights While Preserving Within‑cell Conditionals, and a Noisy‑coin Example Is Used to Compare Numeric Posteriors From Two Updating Routes.

Multivariate information decomposition

Wherein the Mutual Information Held by Multiple Sources About a Target Is Shown to Be Partitioned Into Redundant, Unique, and Synergistic Atoms via a Lattice and Möbius Inversion, While the Original Minimum‑information Redundancy Is Critiqued.

Innovation, science, technology research in Australia

A scrapbook of notes about how research is done in Australia

Wherein Australia’s Research System Is Shown to Be Underfunded at 1.68% of GDP and Burdened by Sprawling Administrative Overheads That Divert Funds and Researchers’ Time.

EAs, rationalists, TPOTs and the like in Australia and surrounding regions

Wherein Networks of Rationalists in Australia and Neighbouring Lands Are Catalogued, and Their Tendency to Be Organised Chiefly via Facebook Groups and Meetups Is Noted, With Local EA and AI‑safety Chapters Listed.

Ensemble Kalman updates are empirical Matheron updates

Wherein the Ensemble Kalman Update Is Presented as an Empirical Matheron Rule, With Perturbed Observations the Stochastic Variant Being Shown to Coincide, and Updates Being Performed in Observation Space So That No D×d Covariance Is Inverted.

Scaling laws for very large neural nets

Theory of trading-off budgets for compute size and data

Wherein Observational Scaling From Nearly a Hundred Public Models Is Described, a Low‑dimensional Capability Space Is Proposed, and RL Scaling Is Shown to Be Far Weaker Than Inference‑time Compute.

Hanging out my shingle

Let’s work on the most urgent thing, together

Wherein a Melbourne Researcher’s Decision to Decline a Decade‑pursued Post Is Announced, and an Offer to Consult on AI Safety, With Availability From Feb 2026 and Mortgage‑driven Runway Noted, Is Tendered.

Parsl

Quiet but deadly Python HPC workflow manager

Wherein Parsl Is Described as a Python‑native Workflow Engine Whose DAG Is Built at Runtime and Whose Slurm Provider Is Shown to Enable Scaling From Laptop to Cluster While Methods for Unwrapping Worker Errors Are Given.

Epidemics and diseases

Wherein Contagion Mechanics and Countermeasures Are Set Forth, and the Proposal That Elastomeric Respirators Be Stockpiled for Critical Workers Is Described as a Physics‑based Defence Against Airborne Threats.

Australia in data

Wherein Australia’s Data Is Surveyed, National Geodata Collections Are Noted, and the AIATSIS Map of Indigenous Australia Is Presented as the Conventional Source for Traditional‑owner Boundaries.

Developmental interpretability

Wherein the Training Evolution of Neural Networks Is Traced, Abrupt Phase Transitions in Loss Geometry Are Examined, Critical Learning Periods Are Identified, Joint Trajectory PCA Across Seeds Is Applied, and Singular Learning Theory Is Invoked.

Normalising flows

Invertible density models, sounding clever by using the word diffeomorphism like a real mathematician

Wherein a Simple Gaussian Base Is Transformed by Composed Invertible Maps, the Reparameterization Trick Is Invoked, and the Need for Tractable Jacobian Determinants Is Noted, Planar and Sylvester Flows Being Given as Illustrations.

Foundation models for geoscience

Wherein a Catalogue of Planetary Foundation Models Is Presented, Their Multi‑temporal Training, Inclusion of Sentinel‑1 Radar and Diverse Spectral Bands Is Noted, and Suitability for H100 Fine‑tuning Is Indicated for Inundation and Burn‑scar Tasks.

Presentation tools

Slide decks, “powerpoints”, beamer lore

Wherein Presentation Tools Are Surveyed in a Systematic Registry, Quarto Is Declared the Preferred Export Path, and HTML Slides Are Noted to Permit Animated GIFs and Offline PDF Export via Decktape.

The denoising diffusion SDE

Stochastic diffusions that are reversible in a computationally useful sense

Wherein the Forward Diffusion Is Described and the Reverse‑time SDE Is Shown to Require the Score, the Gradient of the Log Density, Which Is Approximated by a Time‑conditioned Neural Network Trained by Denoising, Enabling Sampling From Complex Multimodal Targets.

Score matching

Wherein the Data’s Gradient Field Is Inferred by Denoising Noisy Samples With Gaussian Perturbations, So That the Learned Score Is Directed Toward the Expected Clean Sample and Is Recovered as Noise Vanishes.

Coincidences in computation

Formalising “everything happens for a reason” via computational complexity

Wherein a Computational No‑coincidence Conjecture Is Presented and a Neural‑network Null Model Is Developed, With an Advice String and Verifier Being Used to Certify Rare All‑negative Outputs.

Snakemake

Wherein a Build Tool Is Described as a DAG‑driven Workflow Manager for Reproducible Analyses, With Cluster Profiles Used to Submit Jobs to Slurm and Other Schedulers, and Container Support Is Noted.

Feed readers

A standard for user-driven, open news

Wherein the Decline and Revival of Feeds Is Set Out, and a Catalogue of Feed Readers, Self‑hosted Server Options, and Tools to Create Feeds From Sites Without Them Is Presented.

Random features

Random embeddings

Wherein Random, Fixed Cosine Features Are Introduced, Gaussian Frequencies Are Sampled to Approximate the RBF Kernel, and Data Is Projected Into a Higher-Dimensional Space So Linear Methods Are Applied.

Civil society, movements, and AI safety

Movement building, field building, power building, in-fighting in the computationalization of our institutions

Wherein the Nascent Civil‑society Mobilization Around AI Safety Is Depicted as a Movement Ecology Under Strain, With Typical Fracture Lines and Coalition‑dynamics, and Some Relevant Social‑movement Theories Are Sketched.

tmux

Wherein Terminal Sessions Are Managed and Are Persisted Across Logins; Mouse-Driven Scrolling and Clipboard Handling Are Treated as Separate Behaviors, and Plugins Together With iTerm2 Integration Are Noted.

Estimating the Local Learning Coefficient

Singular Learning Theory’s prodigy

Wherein the Local Learning Coefficient Is Estimated by Probing a Hot, Tethered Posterior, Preconditioned SGLD With Inverse Temperature Set to the Reciprocal of Log N and a Gaussian Localizer Is Used, and Implementations Are Provided.

Automatic differentiation

Wherein Automatic Differentiation Is Described via Dual Numbers, Taylor‑series Formulations and Reverse‑mode Backpropagation, and Implementations Such as JAX and Enzyme Are Noted for LLVM‑level and Python Integration.

Material basis of AI

Energy, chips, water

Wherein Economies of Foundation Models Are Examined and the Disproportionate Energy and Water Demands of Large-Scale Training, Including Data‑centre Cooling and Emissions Accounting, Are Described.

Learning to act in generative settings

On the formal theory of choosing to do that which you’ve never seen done before

Wherein Diffusion Models Are Treated as Control Policies, Goal‑conditioned Skills Are Induced by Contrastive Objectives, a Doubly‑optimistic UCB/LCB Algorithm Is Proposed for Paid, Permanent Action Generation, and Empirical Gains on Healthcare QA Are Reported.

Probabilistic programming

Doing statistics using the tools of computer science

Wherein Probabilistic Programming Is Presented as a Means to Express Turing‑complete Generative Models and to Perform Inference, With McMc, Variational Methods and Autodifferentiation Being Supplied.

Research discovery and synthesis

Has someone answered that question I have not worked out how to ask yet?

Wherein the Difficulty of Academic Recommendation Is Stated, and a Retrieval‑augmented Method Is Proposed, Using Tens of Millions of Paper Embeddings to Produce Citation‑backed Literature Synthesis.

Validating and reproducing science

The systems that grind the chaos of noise into good results. Peer review, academic incentives, credentials, publishing…

Wherein the Mechanisms for Validating Scientific Claims Are Catalogued, the Replication Crisis and Publication Bias Are Examined, and Reforms Such as Pre‑registration and Registered Reports Are Outlined.

AI Safety

Getting ready for the grown-ups to arrive

Wherein the Risks of Rapidly Advancing Artificial Intelligence Are Surveyed and Seven Domains Are Enumerated, With Supply‑chain Data Exfiltration and Autonomous Weaponization Singled Out, and Technical Mitigations Sketched.

Scheduling ML jobs on HPC clusters

In Soviet Russia, job puts YOU in queue

Wherein Classic HPC Schedulers Are Surveyed and a Practical Habit Is Recommended: Submitit Is Presented as a Way to Run Python Functions via Slurm, and Snakemake’s Job‑grouping Is Noted.

Making macOS behave itself

Things I have to do to keep my laptop running so I can google how to fix other things

Wherein the Means to Make macOS Behave Itself Are Enumerated, Including an App to Stop Music.app Launching on Device Connection, Terminal Tips for Gatekeeper, OCR From Screen, and Time Machine Control.

Proposal: Modelling and Mitigating Non-Consensual Human Hacking by Advanced AI

Wherein a Formal Theory Is Laid Out, Hypothesizing the Overseer Utility Gap, Deriving Audit-Rate Bounds, and Prescribing a Shaping-Regularizer to Detect and Penalize Non-Consensual Rater Manipulation.

Ensemble Kalman methods

Data Assimilation; Data fusion; Sloppy updates for messy models

Wherein Ensemble Approximations Are Employed to Propagate Low-Rank State Covariances via N-Member Anomalies, and Updates Are Effected in the N−1 Ensemble Subspace Using Perturbed or Square‑root Observation Transforms

ILIAD2

Oddesty

Wherein the Bay Area Unconference Is Recorded, a Neural‑network Analogue of the Computational No‑coincidence Conjecture Is Outlined, and a Phase Transition in Singular Learning Theory Is Noted.

Tolerating Jupyter’s file format

In order to make your python code accessible, we wrapped it into encoded javascript strings. No need to thank us.

Wherein Jupyter’s JSON Notebooks, Swollen With Embedded Binary Media and Outputs, Are Diagnosed as Awkward, and Git-Aware Remedies Such as Nbstripout, Jupytext and Nbdev Are Surveyed.

Jupyter front end systems

UX design by underfunded volunteer committee is how I like my data science

Wherein the Jupyter Ecosystem Is Described as an Ecology of Kernels and Front-Ends, and VS Code Integration Is Noted as an Alternative That Couples Notebook Execution With a Full Code Editor.

Clipboard management

Remembering two things at once

Wherein Clipboard Managers Are Surveyed and the Practice of Syncing Clips — Including OSC52 Terminal Escapes to Push Remote Text Into a Local Clipboard — Is Described, and Attendant Security Risks Are Noted.

Deep linear networks

Let’s pretend our networks are almost polynomial

Wherein the Gradient-Flow Dynamics of Depth-Preserving Linear Nets Are Described, and Singular-Value Trajectories, Mode-by-Mode Learning, and Gated Mixtures Approximating ReLU Are Rendered Analytically Tractable.

Neural flow matching models

Like denoising diffusion except weirder

Wherein Flow Matching Is Presented as a Deterministic Reformulation of Diffusion, Regression on Velocity Fields Is Performed, and Straight‑line Optimal‑transport Trajectories Are Used to Enable One‑ODE Sampling and Exact ICOV Log‑likelihoods.

Conditioning neural denoising diffusion models

Generative modes that match the observations, not the training data

Wherein Conditioning of Neural Denoising Diffusion Models Is Surveyed, and a Twisted SMC Method Is Described That Evaluates the Observation Likelihood at the Denoiser’s Tweedie Estimate of X0 to Guide Particles.

Numerical Python

Wherein Python’s Numerical Ecosystem Is Surveyed, and NumPy Is Noted to Delegate Heavy Linear Algebra to Classic Fortran BLAS/LAPACK While Tools for Printing, Tensors, and Einsum Variants Are Outlined.

Quarto integrated website system

Academic blog publishing that is easy on me, albeit hard on my computer

Wherein the Quarto Integrated Website System Is Described, Its JavaScript‑based Build Pipeline Using Bootstrap, Sass and EJS Is Employed, and a Million‑word Blog Is Handled, Albeit With Lengthy Build and Load Times.

Welcome to cybersec hell

Wherein It Is Recounted How Autonomous AI Agents Have Usurped the Labour of Human Cybercriminals, and How Prediction Markets Furnish a Novel and Troubling Business Model for Their Misdeeds.

AI Safety Australia

Wherein the Australian AI Safety Landscape Is Surveyed, With Senate Inquiries, Public Attitude Research, and a Directory of Local Organisations and Researchers Being Assembled.

Platonic and convergent representations in neural nets

Wherein the Convergence of Internal Representations Across Neural Networks and Human Brains Is Examined, With Attention to Findings That Embedding Models of Differing Architectures May Be Mapped Between One Another Without Paired Data.

What are human values?

If we assume that humans are pursuing stuff, what is that stuff? Thick models of value, eudaimonic rationality,…

Wherein the Limits of Utility Maximisation Are Examined and Thick Models of Value Are Proposed as an Alternative Framework for Understanding Human Motivation and Collective Flourishing.

Mechanistic interpretability

Wherein the Internal Workings of Large Language Models Are Examined Through Reverse Engineering, With Particular Attention Paid to Sparse Autoencoders, Circuit Tracing, and the Contested Notion of Monosemanticity.

Informational empowerment

A particular mathematization of intrinsic motivation

Wherein Informational Empowerment Is Defined as a Mutual‐Information Quantity From Optimal Control Theory, and Its Prospects as an Intrinsic Motivation for Reinforcement Learning Are Examined.

Algorithmic collective action

People, algorithms, coalitions, and the strategic dynamics between them

Wherein User-Collective Leverage Over Platform Retraining Is Set Against the Performative Power of Monopolists, and Stable Agent Coalitions Are Derived via Nucleolus-Based Reinforcement Learning.

Coalition games

Divide-and-rule, formal models for social justice, colonization, elite capture, interest groups, and the like

Wherein the Partition Function Form Is Introduced and Applied to Divide-and-Rule Strategy, With the British Administration of India Examined as an Illustrative Case

Discretizing and quantizing neural nets

In Which Neural Network Quantization Is Surveyed From Its Vector-Quantization Roots to Binary Extremes, With the Role of Calibration Data in Distributing Precision Across Weights Examined.

Medicalization

Wherein the Designation of Human Conditions as Diseases Is Examined, Including Whether Such Categories Constitute Natural Kinds and Why Disease Labels Are Held to Confer Treatment Legitimacy.

AI tutoring

Machines as personalized pedagogues

In Which the Distinction Between Helpfulness and Pedagogy Is Observed, and a Socratic System Prompt Is Applied to Constrain Off-the-Shelf Chatbots Into Reluctant Tutors.

So you’ve joined a union

Field notes for white-collar professionals in Australia

Wherein the Privileges of the Workplace Delegate Are Set Forth, Including Cross-Silo Intelligence Filtered by Managerial Hierarchies, and Enterprise Bargaining’s Bias Toward Incumbent Staff Is Noted.

Nearly sufficient statistics and information bottlenecks

Also, rate distortion by other means, and the measure-valued sufficient statistic

In Which the Posterior Distribution Is Recognised as the Canonical Measure-Valued Sufficient Statistic, Found to Recur Under Many Names Across Disparate Fields, and Subjected to Compression via the Information Bottleneck.

Community sovereign AI compute

On the geopolitical risk of renting your thinking from abroad, and what a small collective can do about it

Wherein the Joint Ownership of AI Inference Hardware by Small Mutual-Aid Collectives Is Examined as a Hedge Against Geopolitical Disruption to Submarine Cables and Commercial API Restriction.

Making good intertemporal decisions

Description TBD

Wherein the Problem of Self-Coordination Across Time Is Cast as an Intrapersonal Game, and Three Classical Strategies — Naïve, Sophisticated, and Resolute — Are Distinguished.

Sovereign compute for small collectives: technical implementation guide

Hardware, software, de-censoring, and the path from cloud audition to owned infrastructure

Wherein the Specifications of a Shared Inference Server Are Laid Out, Its Legal Ownership Is Settled, and the Excision of Political Guardrails From Open-Weight Models Is Treated as Routine.

Computational complexity of Bayesian inference

Wherein the Hardness of Exact and Approximate Inference Is Established, Relative Approximation Is Found to Be #P-Complete, and Tractability Is Shown to Depend Upon Non-Extreme Conditional Probabilities.

DIY Generative AI

Community resource and epistemological infrastructure

In Which the Supply Chain of Open Generative Models Is Decomposed Into Constituent Layers, From Volunteer-Scraped Datasets to Inference Code Running on Consumer Laptops and the Remarkable Facility With Which Communities Are Hacking on the Whole Stack Is Noted.

Civic technology

Utopian governance using technology, including generative AI, Electrohabermas, digital deliberation, platform democracy

In Which AI Mediation of Collective Deliberation Is Examined as an Epistemic Mechanism, With Bridging-Based Aggregation Systems Such as Community Notes and Polis Considered as Means of Surfacing Cross-Partisan Consensus.

Causality, agency, decisions, learning

Causal inference involving agents, decisions, and learning

In Which Influence Diagrams Are Employed as Extensions of Bayesian Networks to Represent Decision Problems, With Multi-Agent Systems and Causal Attribution Also Considered.

Intergenerational game theory

Bargaining with the unborn

Wherein the Structural Absence of Retaliation by Unborn Players Is Examined, and Overlapping-Generations Models Are Considered as Partial Remedies for the Bargaining Gap So Produced.

Game complexity

Wherein the Nash Equilibrium, Whose Existence Is Guaranteed by Fixed-Point Arguments, Is Established as PPAD-complete, While Correlated Equilibrium Is Seen to Admit a Linear-Programming Solution.

Assorted laws and paradoxes

Mostly-eponymous laws I would confound without this note

Wherein an Index of Eponymous Adages Spanning Economics, Computing, and Internet Culture Is Maintained, Several of Which Are Noted to Possess Names Sufficiently Similar as to Occasion Confusion.

Single subject experiments

Instrumentation and analytics for body and soul; Quantified self; precision medicine.

Wherein the Trade Between Internal and External Validity in N-of-1 Trials Is Examined, and Bayesian Methods Are Shown to Dissolve Frequentist Complications Attendant Upon Crossover Designs.

Attention Deficit (Hyperactivity) Disorder

Prescriptive; stimulants, dopamine, nosology and etiology of weird attention spans in modernity

Wherein the Evidence for Stimulant and Non-Stimulant Treatments Is Ranked by Effect Size, and Tolerance to Euphoria Is Distinguished From Tolerance to Therapeutic Benefit in Prolonged Use.

Editing images

Wherein Several Methods of Image Manipulation Are Enumerated, From Command-Line Scripting to GUI Applications, With Particular Attention Paid to the Removal of Identifying Metadata.

Ageing

Prescriptive; vitalism, lifespan, healthspan

Wherein the Evidence for Various Healthspan Interventions Is Surveyed, Daily Exercise Is Found to Dominate All Supplements by a Wide Margin, and Creatine Monohydrate Is Retained on Grounds of Cheapness.

Fashion

Theory and practice of hip garments

Wherein Theoretical Models of Trend Diffusion, Including Red Queen Signalling and Neutral Mutation, Are Surveyed Alongside Vendor Lists for Batik, Bolo Ties, and Antiquated Menswear.

What even is “agency”?

Wherein the Boundaries of Agency Are Explored Through Edge Cases Drawn From Coercive Control, Addiction, Cult Membership, and the Manipulation of Attention and Preference.

What use is utility?

If we must use the expected utility maximizer model for humans, what is the utility we should use?

Wherein the Implied Utility Functions of Animals Are Considered in Light of Machine-Learning Optimisation, and Value Learning Is Introduced as a Framework for Inferring Preferences From Behaviour.

World models arising in foundation models.

Wherein the Convergence of Internal Representations Across Differing Model Architectures Is Examined, and a Fundamental Trade-Off Between Efficiency and Interpretability in World Model Construction Is Revealed.

Causal inference on DAGs

Confounding! This scientist performed a miracle graph surgery intervention and you won’t believe what happened next

Wherein Conditions Under Which Observational Data May Be Used in Place of Randomised Experiments Are Examined, and the Question of Which Variables Are to Be Adjusted For—and Which Are to Be Left Alone—is Resolved.

Demographics, natalism and fertility

Won’t somebody think of the unconceived children?

Wherein the Ethics of Potentially-Existing Persons Are Debated, Parfit’s Repugnant Conclusion Is Encountered, and France Is Noted as the Originator of Birth-Rate Decline Some Centuries Prior.

Games, computer, recreational

Wherein Various Methods of Emulation Are Surveyed, and Browser-Accessible Archives of Vintage Software Are Noted, Alongside Tools for the Construction of New Recreational Diversions.

Melbourne / Naarm

Where even the punks follow rules

Wherein a Newcomer’s Notes Are Compiled on Sparse Wrens, Buruli Ulcers, Sauna Circuits, and Missing‑middle Schemes, With the City’s Trees Being Addressed by Email.

Quantum probability

and quantum information, noncommutative probability

Wherein Classical Real-Valued Chance Is Set as a Special Case of Complex Quantum Measure, and Rao’s Table, Tao’s Free Probability Notes, and Woit’s Free Draft Text Are Noted.

Designing less cruel social media

Wherein an Iterative Remaking of Online Commons Is Proposed, With IndieWeb and Fediverse Noted, and Platform Rules Are to Be Set by Compensated Mini-Publics Convened for Moderation and Privacy.

Voting systems

Mathematics and economics of aggregate group preferences; the will of the people (whatever that might be); social choice theory

Wherein Impossibility Theorems Are Surveyed, Strategic Voting Is Noted, and Quadratic Voting Is Described as Being Funded by Tokens to Register Preference Intensity.

Remote Desktop

Business model: Uber for pixels

Wherein Sundry Networked Desktops Are Surveyed, Protocols Being Noted From SSH X11 Forwarding to NX, VNC, RDP, and SPICE, With Remmina Recorded as a Linux Client, Last Revised in July 2020.

China

Wherein the United Front Work Department Is Observed Recruiting the Diaspora, Clandestine ‘Police Stations’ Being Noted Abroad, and the Great Firewall Being Skirted by Shadowrocket and Shadowsocks.

The levels of simulacra

Slicing up the spectrum of meaning, from literal report to partisan advantage

Wherein the Passage From Plain Report to Political Manoeuvre Is Traced, as Lions Across a River and a China-Borne Pandemic Are Used to Mark Four Truth-Values in Speech.

Causally embedded agency

Embodiment and other origin stories for minds

Wherein Causal Embedding Is Proposed as a Framing for Embodiment and Stochastic Parrotology, and Formal Approaches to Empowerment and an Ecology of Mind Are Promised.

Entropy vs information

MaxEnt(?), macrostates, subjective updating, epistemic randomness, Szilard engines, Gibbs paradox…

Wherein the Relation of Entropy and Information Is Examined, and Macrostates Are Shown to Be Defined as the Unique Markovian Partitions of Phase Space, Linking Thermodynamic Notions to Predictive Causal States.

MaxEnt inference

Looks annoyingly like Bayesian inference, but I’m not convinced

Wherein Constraints Encoding Observations and Prior Knowledge Are Imposed on a Probability Distribution, and the Distribution of Maximum Entropy Subject to Them Is Selected; Connections to Predictive Coding and Optimal Transport Are Noted.

Fun

Homo ludens, play

Wherein the Social Mechanics of Play Are Presented, and the Mining of Creative Resistance by Gentrifying Forces Is Exemplified by Rave Protests, Dance‑floor Politics and Guerrilla Gardening Tactics.

San Francisco Bay Area

Wherein the Bay Area Is Presented as a Magnet for Visionaries and Weirdos, and Its Faltering Public Works Are Documented by Accounts of Sewers, Storm‑water Issues and Ageing PG&E Equipment.

Computation and the edge of chaos

Is criticality what inference looks like?

Wherein an Account Is Given of Computation at Criticality, and the Connection to Deep Neural Networks’ Poised Initialization and Phase‑transition Models Is Examined and Placed Beside Historical Proposals.

Web API automation

Wherein the Art of Web API Automation Is Described as a System of Persistent Change‑checking Services and Numerous Prebuilt Connectors, and Is Noted to Be Implemented via Webhooks and Hosted Functions.

Moloch, slack and friends

Weak selection versus goodharting, 内卷, involution, ascendance, resilience, enshittification, Malthusian equilibria

Wherein the Inevitability of Competitive Selection Is Considered, the Tension Between Ascendancy and Reserve Capacity Is Delineated, and Platform Enshittification Is Exemplified by TikTok’s Progressive Capture of Value.

Is language even symbolic, bro?

Humanistic interpretability

Wherein Language Is Considered as Symbolic and as Performative, Aumann’s Agreement Theorem Is Invoked, Buddhist Koans Are Noted, and Post‑symbolic Communication À La Lanier Is Sketched.

Probability

Wherein an Account Is Presented of Probability as Models for Event Regularities, Founded on Conditional Probability and Rényi and Cox Formulations, and an Introductory Animated Lecture Is Noted.

Hygienic masks

Medical/anti-infectious masks

Wherein the Reader Is Informed That Elastomeric Respirators With P100 (P3) Filters Are Recommended as Reusable, Better‑fitting Alternatives to Disposables, and Simple Fit Hacks Are Described.

Morality and computational constraints

It is as if we knew what we were doing

Wherein Computational Constraints and Moral Theory Are Examined, and the Role of Reinforcement‑learning Reward in Signaling Pain and Workplace Health‑and‑safety Risks Is Considered.

Pornography and other lewd art

Morsels of oddity from the depiction of human sexual behaviour

Wherein a Miscellany of Writings on Pornography Is Presented, and Attention Is Drawn to Feminist Video Porn on PinkLabel and the Emergence of AI‑generated Marketplaces Where Anyone’s Image Is Commodified.

Comfy Arch Linux

Wherein the Rolling, Self‑assembled Arch Is Contrasted With Staged Updates and GUI Tooling in Manjaro, and With CachyOS’s CPU‑tuned Kernels and Performance Defaults, While Update Disruptions Are Noted.

ELBO

Evidence lower bound, variational free energy

Wherein the Evidence Lower Bound Is Presented as a Decomposition Into an Approximating KL Term and an Expected Log‑likelihood Plus Entropy, and Importance‑weighted Sampling Is Indicated as Next Step.

Functional programming

Wherein Computation Is Treated as the Evaluation of Mathematical Functions, Mutable State Is Avoided, Functions Are Allowed as Values, and Their Use in Differentiable and Probabilistic Languages Is Noted While Memory Reuse Is Studied.

Python caches

The fastest code is the code you don’t run, especially python code

Wherein a Survey of Python Caches Is Presented, and Disk‑backed, Multiprocess‑safe Solutions That Store Large Binary Blobs Without a Server Process and With Minimal Installation Are Examined.

Gradient descent at scale

Practical implementation of large optimisations