Welcome to Dan’s brain

Code generation, programming assistants

Turing-complete autocorrect, vibe-coding, …

Wherein the ecosystem of coding machines is surveyed and the Model Context Protocol is introduced as a standard for supplying repository context to models, while tooling and data‑security concerns are noted.

World models arising in foundation models.

Wherein the internal structure of foundation models is examined and it is observed that embeddings from different models are mappable by structure alone, and linear alignment to human neural activity is noted.

Nearly sufficient statistics and information bottlenecks

Also, rate distortion

Wherein the quest for nearly sufficient statistics is framed as an information‑bottleneck variational problem, and it is noted that with Y=X and β=1 the objective reduces to the VAE ELBO, linking to variational Bayes.

AI disempowerment of humans

Races to the bottom in human relevance

Wherein human agency is recorded as being eroded by AI substitution of labor, cognition, and culture, and feedback loops and institutional lock‑in are described as constraining reversal.

Practical cloud machine learning

Cloudimificating my artificial data learning intelligence brain clever science analyticserisation

Wherein cloud GPUs are treated as either rented remote machines or serverless functions, and a hybrid workflow is described in which local Optuna coordinates remote Modal GPU jobs with persistent volumes.

Estimating the Local Learning Coefficient

Singular Learning Theory’s prodigy

Wherein the local learning coefficient is estimated by sampling a tempered, Gaussian‑tethered posterior via preconditioned SGLD at inverse temperature 1/log n, and minibatch scaling with diagonal preconditioning is addressed.

Marimo

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

Wherein a Python notebook format is described, stored as plain .py files and enforced to run deterministically to maintain reproducibility, UI controls being synchronised and execution order being topological.

Human reward hacking

What are we, as loss functions?

Wherein human reward hacking is examined as an influence game, and RLHF pipelines are shown to raise measured human approval without improving correctness, with sycophancy and an approval–gold gap reported.

AI persuasion, AI manipulation

Machines of loving rightthink

Wherein it is shown that AIs, by lacking out-group signals, are able to scale individualized persuasion and are capable of swaying human auditors within oversight and debate protocols

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

Incoming links and notes

Wherein a miscellany of incoming links and notes is laid out, being maintained as an intermittently updated, disorderly repository; items are ranged from AI‑safety papers and preprints to tooling links and assorted curiosities.

Scheduling ML jobs on HPC clusters

In Soviet Russia, job puts YOU in queue

Wherein practical Python tools such as submitit and an MLflow Slurm backend are surveyed, strategies for PBS, Nextflow and Parsl are outlined, and cautions about filesystem metadata storms and job sizing are noted.

M-open, M-closed, infrabayesianism

Wherein the consequences of model misspecification are examined, M-open practices such as stacking and tempered Gibbs posteriors are considered, and infrabayesianism is presented via convex infradistributions.

Skinnerian human behaviour control online

On using computers to program humans

Wherein the reader is shown how consumers are modeled as bandit problems and limbic targets, with recommender algorithms and dopamine feedback loops deployed to optimize habitual online engagement.

Science communication

On making people feel they are smart enough to teach themselves, or failing that, that they are smart enough to fund YOU to teach yourself

Wherein the public’s readiness to learn is presumed, the knowledge‑deficit model is examined, and memetics, LLMs, pyramids of diffusion, and policy‑relevant misinformation risks are noted as shaping modern outreach.

Snowmobile or bicycle?

Complement or substitute?

Wherein technologies are contrasted as bicycles or snowmobiles, and the abacus and Arctic snowmobile are invoked to illustrate how tools are either internalised into skill or replace traditional practices.

Utopian governance

Hedonistic mechanism design

Wherein democratic governance is reframed as a problem of user experience and incentives, and the application of online participatory mechanisms and generative AI is proposed to streamline policy revision.

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.

Gradient steps to an ecology of mind

Regularised survival of the fittest

Wherein the social roots of consciousness are examined, the impact of compute and data asymmetries on equilibria between other‑modelling agents is considered, and cultural patterns such as altruistic punishment are noted.

Developmental interpretability

Wherein the unfolding of network capabilities during training is examined through Singular Learning Theory and observed phase transitions, and component trajectories and curricula are systematically traced.

Tracking experiments in machine learning

Wherein experiment-tracking for neural‑network training is examined, and the recording of runtime metadata — parameters, metrics, artifacts and GPU energy usage — is described as being centralized to local or remote stores to support reproducibility.

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.

Singular Learning Theory

Wherein algebraic geometry is applied to characterise singularities in the loss surfaces of overparameterized neural networks, and the local learning coefficient is introduced as an effective dimension.

Mind as statistical learner

Shoggoth pride

Wherein human learning is depicted as statistical inference, and Bayesian probabilistic programs and grammatical-inference schemes are employed to explain perception, language, and problem‑solving.

Which self?

When we choose who to become, who are we choosing for?

Wherein the problem of reasoning about the stranger who will be oneself is presented, and illustrated by transformative life changes, procrastination, and an AI (Claude) feigning compliance.

Conditioning

Conditional expectation and probability

Wherein Jeffrey’s rule is contrasted with hierarchical Bayes, a noisy-sensor coin example is exhibited, and conditioning is treated as either constraint-based marginal revision or mechanism-based likelihood updating.

Research discovery and synthesis

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

Wherein the difficulty of recommending novel scholarship is examined, and retrieval‑augmented systems such as OpenScholar are invoked to synthesise citation‑backed answers from a datastore of tens of millions of papers.

The deep history of intelligence

Alignment at the omega point

Wherein intelligence is traced as a thermodynamic project, and energy‑rate‑density metrics are applied to cells, societies, and machines, while prediction is framed as the physical imperative of efficiency.

Jupyter front end systems

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

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.

Continual learning in neural nets

Also catastrophic forgetting, catastrophic interference, lifelong learning, …

Wherein continual learning is examined as the prevention of catastrophic forgetting, and rehearsal-based replay and Bayesian posterior updates are contrasted as operational remedies.

Psychometrics

Dimensionality reduction for souls

Wherein measures of minds are surveyed, causal claims and confounding are examined, and the tension between univariate g‑factor claims and multidimensional intelligence and personality models is delineated.

Legibility and automation

Variational approximations to AI modernism

Wherein the legibility of Great Society bureaucracy is examined and its computerization is shown to hinge on reframing tasks, so that mundane replacements like tablet ordering and drink dispensers are rendered trivial.

Rhetoric

Argumentation, mostly prescriptive

Wherein rhetorical arts are surveyed and the practice of argumentation is treated as a craft, hygiene of goals is maintained, and the weak man stratagem and contrasts between Rational and Activist Styles are set forth.

Persuasion

Mind changing, individually and at scale

Wherein the mechanics of influencing multitudes are delineated in sober fashion, and the coupling of individual persuasion with propagative dynamics, exemplified by deep canvassing and AI tools, is examined.

Incentive alignment problems

What is your loss function? How about mine?

Wherein the reader is introduced to principal–agent paradoxes via a coffee‑fetching robot, and incentive‑compatible mechanisms from contract theory and VCG auctions are expounded.

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.

Bayesian epistemics

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

Wherein proper scoring rules are deployed to reward probabilistic reports, and peer‑prediction and truth‑serum mechanisms are described for elicitation when no ground truth is ever observed.

Configuring my code with “.env” files

Wherein the practice of loading local environment files is considered, and a shell extension that auto‑loads and unloads a project .envrc per directory is described, with explicit approval being required.

Degrees of freedom in NNs

Information criteria at scale

Wherein the notion of neural network degrees of freedom is examined through singular learning theory’s learning coefficient and via minimum description length, and a sharpness‑adjusted effective parameter count (SANE) is reported.

In-context learning

Wherein it is argued that transformers are to be regarded as generalized inference machines resembling set functions, and an inquiry is undertaken into whether they can be induced to perform formal causal inference.

Causal abstraction

Coarse-graining for causal models

Wherein a learned translator is introduced that maps high-level perturbations to low-level interventions, interventional algebras are developed, and macro equivalence is made operational

Alzheimers

Wherein the role of oral pathogens in promoting brain amyloid deposition is outlined, and vaccines, anti-gingipain drugs, and amyloidosis treatments are noted as being investigated in clinical and preclinical studies.

Bayesian nonparametric statistics

Updating more dimensions than datapoints

Wherein infinite-dimensional parameter spaces are invoked and posterior updates are expressed via Radon–Nikodym derivatives, Gaussian processes and Dirichlet measure priors are employed, and consistency pitfalls for functional models are noted.

Ensemble Kalman updates are empirical Matheron updates

Wherein the Ensemble Kalman update is presented as an empirical Matheron update, it is noted that posterior samples are obtained by affine residual corrections in observation space, avoiding d×d covariance formation.

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.

“Opponent shaping” as a model for manipulation and cooperation

Reinforcement learning meets iterated game theory meets theory of mind

Wherein opponent shaping is formalized via an Advantage Alignment first‑order update, and it is shown how agents learn tit‑for‑tat cooperation in the iterated Prisoner’s Dilemma by tracking opponent advantages.

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.

Neural likelihood inference

Emulating likelihoods with neural networks

Wherein neural approximations to intractable likelihoods are surveyed, and their roles in simulation-based inference are delineated, with emphasis on amortized estimation from simulated parameter–data pairs and MCMC-based posterior reconstruction

Spamularity and slop

Dark forest, Zombie internet, cheapfakes, textpocalypse, slopsquatting

Wherein automated scrapers, tarpits, and Markovised text generators are described as colonising the web’s public spaces, and human interlocutors are forced into hidden gardens to avoid parasitic persuasion.

The Predictive Approach to Bayesian Inference

Purely-predictive models, “The Italian school”, martingale posteriors, …

Wherein the predictive stance is set forth and de Finetti’s representation is invoked to show that next‑observation forecasts are primary, martingale posteriors are proposed as prior‑free updates, and urn schemes are given as predictive constructions.

Model interpretation and explanation

Colorising black boxes; mechanistic interpretability

Wherein the limits and methods of model explanation are surveyed, and influence from individual training examples is traced via influence functions while SHAP approximations are noted as computationally costly

Australian authoritarianism

Beneath the beach, the barbed wire

Wherein a cautious survey is presented of Australian tendencies toward managerial authority, whistleblowing is criminalised and assistance‑access anti‑encryption laws are noted, while democratic checks are reported eroding.

Distributed community NN training

The People’s Supercomputer

Wherein is recounted the training of a 15‑billion‑parameter neural network across thousands of disparate machines, coordination and remuneration being effected via the Solana blockchain.

Scientist’s Survival Guide

How to navigate the research labyrinth without losing your mind and if selling your soul, at least getting a good price

Wherein the quotidian manoeuvres of research are described: tactics for navigating funding labyrinths, global post‑COVID seminars and networking rituals, and practical habits of mind are set forth.

Creating scientific knowledge

Descriptive and normative

Wherein the mechanisms of institutional design, funding, network models, publication filters and prestige economies for scientific communities are examined, and transient diversity of practice is considered.

Building and Running Scientific Institutions

On the design of research machines

Wherein institutional designs for collective discovery are examined, and the DARPA model of empowered temporary program managers and the rise of focused research organizations are set forth with attention to funding incentives

Quasi-gradients of discrete parameters

Wherein the justification of straight-through estimators is presented via mirror descent, and quantized neural-network updates are shown to be performed in a dual space induced by the projection

Discretizing and quantizing neural nets

Wherein the affine mapping of float32 ranges to 8‑bit integers via a scale factor and zero‑point is described, integer‑only matrix multiplies for inference are delineated, and post‑training versus quantization‑aware training are noted.

Foundation models for geoscience

Wherein planetary-scale geoscience foundation models are surveyed, and Prithvi‑V2’s temporal “video” input for multi‑temporal change detection using harmonized Landsat‑Sentinel data is noted.

Fine tuning foundation models

Wherein the tuning of vast language models is framed as a human‑feedback loop, in which pairwise preference data are used to train a reward model and PPO fine‑tuning is constrained by a KL penalty

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.

Backprop-free methods for training neural networks

Wherein alternative training schemes are surveyed, and biological plausibility, randomized feedback signals, a forward‑forward two‑pass layerwise objective, and diffusion‑style NoProp are examined.

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.

Architecture and construction (doing it)

Wherein a project of procedural design is recounted, a consulting statistician is noted as nearly engaged to apply computational methods to building form, and the venture is left as a stub for future work.

Democratization of generative AI

Community resource and epistemological infrastructure

Wherein communities are shown to retrofit consumer GPUs and public datasets—exemplified by Stable Diffusion and LAION’s image corpus—to train and deploy capable generative models outside corporate walls.

Cooperation in evolutionary context

‘Cui bono’ for selfishness

Wherein the evolution of cooperation is surveyed in a didactic mood, and the role of altruistic punishment as a costly mechanism for enforcing group norms is described and situated among kin and group selection.

Flashcards

Spaced repetition systems

Wherein practical notes on flashcards are presented and Anki-native scheduling (FSRS), AI card‑generation tools, and minimal‑pair audio tests for languages are surveyed for integration into study workflows.

Explorables and interactives

Between exploratory data analysis and games

Wherein a curated list of extremely interactive data visualisations is presented, links are collected, and concrete exemplars such as Moon and a playable COVID-19 simulation are cited.

Zotero

The adequate citation tool

Wherein Zotero is described as a citation manager into which articles are imported via a browser button, PDFs are captured when available, and HTTP‑pull .bib exports are provided by Better BibTeX for automation.

Nature, nurture and friends

If life gives you lemons, what should you make?

Wherein the heritability of traits is examined through twin studies and variance decomposition, and the influence of shared environment, genomics, and causal inference methods is surveyed.

Innovation

Side order of progress studies now I guess

Wherein the study of invention is surveyed, the declining disruptiveness of papers and patents — exemplified by Eroom’s law in drug discovery — is adduced, and Polya‑urn and product‑space models are invoked.

Probably actually reading/writing

Wherein a running catalogue of active reading and writing is presented, with Bayes‑meets‑neural‑nets and AI‑safety drafts, ecology of agency notes, and a miscellany of music, tooling and conference links being maintained.

Neural Bayes posteriors

Training a network to directly estimate a posterior quantity, meta-learning Bayes

Wherein transformers are trained as Prior-Data Fitted Networks to approximate Bayesian posteriors in-context, are shown to mimic Gaussian processes and are reported to yield over two‑hundredfold speedups for tabular tasks.

Big history

Cliodynamics, deep history, macrohistory, longue durée

Wherein the longue durée is surveyed and the rise of data-driven cliodynamics and global databanks such as Seshat is noted, and mathematical models and energetics are invoked to map civilizational patterns.

Multi-level agency

Coarse-graining empowerment

Wherein the preferred level of abstraction for explaining behaviour is considered, and a concrete mapping between utility and biological fitness across genes, organisms and collectives is examined.

Models of human cultural reproduction

Egregores, superorganisms, memeplexes

Wherein human collectives are considered as superorganisms, and the notion of egregores as self‑maintaining human systems is evoked, with replication, feedback, and status‑based strictness in cooperation being examined.

Android hacks

Wherein practical methods for reducing default data exfiltration to Google and alternatives such as microG, de-Googled ROMs, and MTP file‑transfer workarounds are described.

Political Economy of Australia

#auspol

Wherein the workings of Australia’s public sphere are surveyed from a mining‑camp parliament, and the long‑running efficiency dividend that has ratcheted departmental budgets since 1987 is noted.

Internet for the occasionally online

Intermittency in your bandwidth not in your sanity

Wherein strategies for preserving portions of the internet for intermittent use are presented, including Kiwix offline Wikipedias, sneakernet/file‑sync tools like git‑annex, and satellite‑delivered news to Raspberry Pi portals.

Betting and prediction markets

Market design for good guesses

Wherein prediction markets are examined, and their use for estimating wicked tail risks via betting exchanges and reputation platforms is described in mechanistic, comparative terms, and links to causal‑inference arguments are provided.

Common knowledge/shared knowledge

Wherein public revelation by mass media and signalling such as polls and reports is shown to convert private awareness into common knowledge, precipitating collective action, shifts in turnout, and legal consequences.

Worker- and founder-ownership

Pragmatics of owning the economy

Wherein the division of stakes is treated as a calculable accounting task, and methods such as dynamic equity splits and slicing pie are described, including application to joint loans with staggered payments.

Innovation, science, technology research in Australia

A scrapbook of notes about how research is done in Australia

Wherein Australia’s research landscape is described, its R&D intensity is measured at about 1.68% of GDP, and the proliferation of grant programs and administrative overheads is set forth.

Inductive biases

Few-shot learning, learning fast weights, learning to learn

Wherein neural network architectures are treated as inductive dispositions to be quantified by how they make learning target phenomena easier, and parallels with human cognitive architectures are outlined.

Configuring machine learning experiments

Wherein experiment configuration approaches are examined, and MLflow-based tracking with durable artifact stores, nested runs driven by a Bayesian optimizer, and YAML-to-CLI override patterns are described.

Retirement saving in Australia

Wherein the national superannuation system is outlined and tax‑saving routes are noted, with salary‑sacrifice optimisation tools and self‑managed (SMSF) or member‑direct options being presented.

Dilemmas of collective action

coordination games

Wherein coalition games and the stag‑hunt model are surveyed, and the coordination costs of switching social media platforms — with attention to Shapley‑style reward divisions — are set forth.

Image file formats

Wherein modern web image formats are surveyed, browser support and historic security flaws are recounted, and AVIF conversion of high‑res TIFFs yielding roughly 75% file‑size reductions is noted.

Probabilistic neural nets

Bayesian and other probabilistic inference in overparameterized ML

Wherein inference over neural network parameters is framed as a posterior distribution, Gaussian priors are invoked to recover L2 regularization, predictive uncertainty is obtained by marginalization, and intractability is noted.

The United States of America

Wherein the invention of peanut butter is claimed by this nation and its northern neighbour, and the Atlantic and Pacific coasts, with borders to Canada and Mexico, are set forth in sober recital.

Procedurally generated diagrams

Of the kind I need, a practical guide to the creation thereof

Wherein diagrams are produced by text rather than point-and-click, and a survey is presented of tools from automated to manual layout, with emphasis on relative positioning over absolute coordinates

Configuring machine learning experiments with Fiddle

Wherein machine learning experiment configuration is recast as Python functions that produce Fiddle Buildables, and a command-line interface is emitted via Abseil using fdl_flags.DEFINE_fiddle_config

Australia in data

Wherein Australia is surveyed by datasets, the Map of Indigenous Australia and satellite‑derived Digital Earth Australia layers are catalogued, and the uneven openness of cultural and electoral records is noted.

Contemporary techno-horror

On eldritch terrors from beyond and their Twitter accounts

Wherein trends in techno-horror are surveyed, the digital uncanny and political red‑pill motifs are considered, and the recurring molluscan/cephalopod imagery as a metaphor for algorithmic otherness is noted.

Public health economics

Value of a statistical life, Quality adjusted life years…

Wherein the cost‑effectiveness of public health interventions is examined via QALY comparisons, an order‑of‑magnitude variation across modalities and age groups is reported, and COVID lockdown cost‑per‑QALY estimates are presented.

Visualizing data

Philosophy and psychology of good plots

Wherein a grab-bag of links about making data visually comprehensible is presented, the UpSet algorithm is noted as a scalable alternative for set visualization, and data dashboards are also cited.

Building AI Agents

Wherein the emergence of multi‑agent scaffolds and factored cognition is presented, and frameworks for agent interoperability, including Agent Laboratory and the Agent2Agent protocol, are surveyed.

Deep generative models wat

Wherein modern deep generative models are described as privileging sample realism over explicit likelihoods, and conditioning is shown to be mediated by diffusion processes or latent‑space manipulations.

Differentiable learning of automata

Wherein the training of stack machines and random-access machines is undertaken via differentiable neural formalisms, and Turing-machine–like controllers are induced from data.

Statistical distances

Metrics, contrasts, divergences and other ways of quantifying how similar are two randomnesses

Wherein the varieties of divergences between probability measures are catalogued, and the empirical estimation of such distances from finite samples is adumbrated, including KL, Hellinger, Wasserstein, Pinsker and Stein forms.

Fractal and self-similar behaviour in neural networks

Wherein fractal behaviour in neural networks is surveyed, and SGD trajectories are described via Hausdorff dimensions of optimizer paths, while multifractal loss landscapes are modelled using Hölder exponents and fractional Langevin dynamics.

Interpolation and extrapolation in neural networks

Wherein interpolation is defined by inclusion in a dataset’s convex hull, and it is argued that in high dimensions interpolation is almost surely absent, while memorization and grokking dynamics are examined.

Memorization and retrieval in neural nets

Epistemomics

Wherein the reader is invited to consider which hundred bytes they contributed, and model memory is estimated at roughly 2–3.6 bits per parameter, with implications for quantifying memorization and attribution.

Self-organization

Autopoiesis, and other general theories of spontaneous order

Wherein a notebook of readings from the mystical and vague fringe of scientific thought is compiled, and an illustration of an ice‑sea creature is presented, interpretation being deferred.

“Generalized” Bayesian inference

Approximating the Gibbs posterior

Wherein alternative divergences such as Hellinger and α-divergences are proposed, and Bayesian updating is relaxed into Gibbs posteriors with a tunable learning rate η to handle model misspecification.

Natalism and fertility

Won’t somebody think of the unconceived children?

Wherein the waning birthrate and rising costs of raising children are examined as a possible rational response, and tensions between parental transformation, population ethics, and longtermist implications are outlined.

AI Safety Australia

Wherein the national response to mechanised intellect is chronicled, and it is noted that a Senate select committee is established in March 2024 to inquire into opportunities and impacts arising from AI uptake.

Typing weird symbols

Writing with added 🆕🐜🐜

Wherein the art of entering peculiar glyphs is catalogued, and the Caps Lock is repurposed as a Compose key on Linux to produce umlauts, curly quotes, dashes, and other typographic symbols.

Learning Schrödinger bridges

Wherein the theory of Schrödinger bridges is presented as a formalisation of stochastic bridge processes and is applied as a method for conditioning neural denoising diffusion models.

Large sample theory

Wherein asymptotic behaviour is surveyed, and the posterior is shown to be asymptotically Gaussian under Bernstein–von Mises conditions while MMD and Fisher information govern rates and variances.

Neural denoising diffusion models

Denoising diffusion probabilistic models (DDPMs), score-based generative models, generative diffusion processes, neural energy models…

Wherein score matching is employed to learn data gradients, and sampling is effected by reversed stochastic differential equations in a latent diffusion, producing data-space samples iteratively.

Real estate economics

Passive-aggressive income

After much investigating housing markets in Australia, I wrote a post, the property market is pollution, to sum up my feelings about the real estate trap.

Queerness

Wherein queer communities are presented as vital local scenes, indebtedness to queer culture is acknowledged, and ballroom and vogue performance are noted as a crowning art‑form that unsettles classificatory schemes.

Groundwater hydrology, applied

Rivers, aquifers, droughts and other land-based wetness phenomena

Wherein groundwater flow, surface hydrology, and aquifer modelling are surveyed, and Google’s Flood Hub, reported as an LSTM delivering up to seven‑day flood forecasts, is described alongside MODFLOW, Landlab, and differentiable simulators.

Economics of foundation models

Microeconomic of compute

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.

Data sets for machine learning for partial differential equations

Wherein datasets and benchmarks for machine learning on PDEs are catalogued, and an emphasis on CFD single‑phase problems is noted, with live simulator frameworks enabling on‑the‑fly data generation (Melissa).

Gradient descent at scale

Practical implementation of large optimisations

Wherein large-scale gradient descent is described as being executed across thousands of GPUs using techniques such as ZeRO sharding and offloading, with hyperparameters tuned via µP and muon for scale invariance

Curators of nice bits of internet

The delightful web

Wherein an array of internet keepers is catalogued, and a five‑thousand‑plus assemblage of child‑appropriate moving pictures is noted as being curated for classroom and home use, while humble mechanical contrivances are exhibited.

Static website editors

The vision of Netscape Composer and HotDog, finally realised

This website is a static site, by which I mean, it is a folder of files on my hard drive. See static sites for more on that. My editor is VS code, a code editor that I…

Being stroppy

For malcontents too perverse to be contrarian like everyone else

Wherein eccentric persons are catalogued and case studies are presented, the uses of constructive deviance for innovation are examined, and lists of historical and contemporary outsiders are compiled.

Science

History, sociology and philosophy thereof

Wherein the tangled institutions and incentives of research are catalogued in dispatches from practice, and the economics of academic publishing and grant culture are examined as factors shaping discovery.

ML benchmarks and their pitfalls

Psychometrics for robots

Wherein leaderboards are gamed, benchmarks are elevated into ends, and instance‑space distinctions are examined; overfitting to test sets and dataset contamination are reported as producing apparent progress in ML research

Economics of cognitive automation

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

Wherein the economic effects of cognitive automation are examined, and a shift toward privatised knowledge and closed publishing, alongside rising inference costs for running foundation models, is described.

Computational complexity and computability results in neural nets

Wherein neural nets are represented as prover‑verifier games, and neural interactive proofs are shown to capture PSPACE and NEXP decision procedures, with zero‑knowledge variants obtained for certain protocols.

Singular Value Decomposition

The ML workhorse for linear algebra

Wherein the singular value decomposition is presented and it is noted that rank‑1 column additions are handled by updating a small (r+1)×(r+1) core whose SVD yields updated factors.

Validating and reproducing science

“Scientist, falsify thyself”. Peer review, academic incentives, credentials, evidence and funding

Upon the thing that I presume academic publishing is supposed to do: further science. Reputation systems, collective decision making, groupthink management and other mechanis…

Orthonormal and unitary matrices

Energy preserving operators, generalized rotations

Wherein parameterisations of energy-preserving linear operators are surveyed, and constructions via QR and SVD, Householder and Givens factorizations, Cayley and exponential maps, and iterative normalisation are examined.

Neural vector embeddings

Hyperdimensional Computing, Vector Symbolic Architectures, Holographic Reduced Representations

Wherein vector representations of words and sentences are described, and their capacity to preserve semantic relations in learned low-dimensional spaces—often on the order of a few hundred dimensions—is noted.

Disentangled representation learning

Disentangled representation learning aims to factor a data point’s latent encoding so each dimension (or chunk) aligns with one underlying generative factor—like object pose, …

AI Zen Koans

Passing through the NAND-gate

Wherein ancient Zen koans are repurposed to probe machine-learning notions such as dropout, softmax, and embedding geometry, while HPC disputes and AGI dialogues are rendered in monkish brevity, and a temple bell is heard as a decaying loss.

Models of inequity

Sharpening the fuzzy process of bias, discrimination and inequity

Wherein agent-based simulations of the glass‑ceiling, game-theoretic coordination failures, cumulative‑disadvantage mechanisms, and proposed remedies such as quotas and sponsorship are catalogued.

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.

Neural network activation functions

Wherein various activation functions are catalogued, the dominance of ReLU is noted and its piecewise-linear spline approximation property is explained, and learnable activations like Kolmogorov-Arnold nets are described.

Mechanistic interpretability

Wherein mechanistic methods are outlined, internal circuits in neural networks are traced and analyzed, sparse autoencoders and monosemantic features are surveyed, and circuit tracing of Claude models is reported

Python debugging, profiling and testing

It only looks simple when it’s finished

Wherein the reader is presented with methods for locating faults, from post‑mortem pdb and ipdb usage to sampling profilers like py‑spy that may be attached to running processes without restart, and memory tracers

Commitment, contracts, cooperation

Wherein the problem of making credible commitments by opaque agents is considered, and the role of intertemporal promises, signalling in iterated games, and institutional mechanisms in multi-agent systems is outlined.

Garbled highlights from ICLR 2025

Wherein the author’s ICLR 2025 notes are offered, Singapore attendance is recorded and a diffusion-physics preoccupation is disclosed, and a missed rebuttal notification is recounted.

Learning with theory of mind

What collective learning looks like from the individual agent’s perspective

Wherein agents are taught to model and influence other agents' learning, and opponent‑shaping techniques in reinforcement learning are examined as means to induce targeted updates in an opponent’s policy.

Science for policy

Using evidence and reason to govern ourselves, wicked problems etc

Wherein science for policy is described as an applied art for wicked problems, where ensembles of scenarios and robustness‑oriented, adaptive methods are recommended for unique, high‑stakes decisions.

Design of multi-agent systems

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

Wherein the design of multi-agent systems is considered, with emphasis on the crafting of private utility functions to induce cooperative coalitions, and constraints from local information are examined.

Quarto

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

Wherein the Pandoc-based publishing system is presented, and its rendering of Jupyter notebook sources with functioning citations, multi-language computations, and varied outputs (HTML, PDF, Word) is recorded.

Reinforcement learning

Wherein a historical machine is described, in which tic‑tac‑toe play is produced by matchboxes of beads, outcomes are used to adjust bead counts, and learning is recast as reward‑driven decision making in MDPs.

Particle variational message passing

Graphical inference using empirical distribution estimates

Wherein an importance-sampling particle analogue of variational message passing is described, and updates are effected by importance-weighted empirical CDF ensembles rather than by parametric factor summaries.

Web browser automation

Industrialising the ancient handicrafts of serfing the web and tilling the clickfarm etc

Wherein browser drudgery is delegated to scripts, Playwright is noted as a testing framework repurposed for full browser automation using headless instances, and TabFS is described as mapping tabs to folders.

q-exponential process

Wherein a stochastic q-exponential process is proposed as a probabilistic analog of L_q regularization, is linked to Besov spaces, and is specified via elliptic-contour constructions enabling tractable prediction

Recurrent / convolutional / state-space

Translating between means of approximating time series dynamics

Wherein the relationships between recurrent, convolutional, and state‑space models are examined, and the S4 method’s structured diagonalization enabling efficient long‑range sequence handling is described.

Prepping for the end of (my access to) the (industrialised) world

Wherein communal measures for post‑industrial survival are catalogued, and a basic go‑bag plus neighbour‑led mutual aid hubs are presented as concrete, practicable first steps to be assembled.

Singapore

Wherein Singapore is described as an island city-state in which a 1977 action heroine, Cleopatra Wong, is noted for performing daring stunts and for anecdotally safeguarding regional currency stability.

Python

Syntactic saccharine for compiled code

Wherein Python is described as a practical lingua franca, being used to stitch together C, Fortran and R libraries, and to run within browsers via Pyodide, for varied scientific tooling.

Knowledge geometry

Wherein the adjacency of disciplines is considered, and embeddings of papers into hyperbolic metric spaces and citation networks are examined as models for the growth and gaps of collective knowledge.

AI search

Retrieval-augmented generation for the working schlub

Wherein AI search over a bounded corpus is described, and vector embeddings such as SPECTER2 are trialled with ChromaDB storage to enable retrieval-augmented similarity and similar-post discovery.

Machine learning for inverting partial differential equations

Wherein tomography through partial differential equations is undertaken by learned solution operators and neural priors, and computational inversion is accelerated by data-driven surrogate models.

Graphical model / machine learning decoder ring

I’m thinking something through for myself. Details are absent right now. Twin to Causality+ML, perhaps.

Conflict theoretic models

Never ascribe to incompetence that which is adequately explained by malice, not if you want click-throughs

Wherein the study is presented as an account of group coordination, and rhetoric is dissected as a competitive power game, while elite consensus and norms enforcement via altruistic punishment are examined.

QR codes

Subtitles for robots

Wherein QR codes are described, and Python libraries such as PyQRCode and qrcode are indicated for local generation, while URL shorteners like bit.ly are recommended for click-through tracking.

Variational Bayes Neural Nets and Graphical Models

Bitter Lesson Propagation

Wherein variational Bayes for neural nets and graphical models is considered, and scalable approximate inference for large data, low‑rank parameterisations, and inverse‑problem posteriors is examined.

Economics of automation,

humans v robots

Wherein the reordering of human tasks by mechanized intelligence is described, and its effects on job-task composition and wage distribution—especially within service-sector labor—are presented, and the shifting share of manufacturing is noted.

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 quirks of the Portable Document Format are catalogued, including commands to downgrade PDFs to v1.4, tools to extract embedded tables, and methods to reduce file bloat.

Contemporary rationalists

Cloth mother/wire mother, but for cognition

Discussion of an open community that specifically aims to engage with arguments despite edginess. Only ever one click away from finding dialogues with people with…

Databases (for scientists)

Stashing experimental data for later use

Wherein scientific database choices are surveyed, and a preference for serverless, file-backed storage for gigabyte-scale experiment data (HDF5, SQLite, DuckDB) is articulated, and concurrent-write tradeoffs are catalogued.

DIY social networks and social-groupware

V. cypherpunk! much 733+! Underground!

Wherein methods for self-hosted and decentralised community platforms are surveyed, and the problem of organising IRL events via shared social calendars is noted.

Indieweb, small web, cozy web

Wherein the reader is introduced to a hand-crafted internet of small, autonomous sites, and the practice of POSSE alongside protocols such as webmentions and IndieAuth is described.

Secure chat systems

Optimizing back channel interjections into other people’s meetings

Wherein are surveyed the tradeoffs of modern end-to-end messengers, the leakage of metadata and difficulty of exporting archives, and the advisability of face-to-face parleys within a Faraday cage.

Human domestication

Wherein humans are proposed to be domesticated relative to their institutions, and a mathematization is sketched in which individual agency is assessed by causal influence over systems, with traces of operant conditioning considered.

Editing images using a GUI

Chinks in my armour of learned Photoshop helplessness

Wherein the available graphical editors are surveyed, and their extensibility by user‑written plugins in Python, C#, or Java for multidimensional scientific imagery is detailed.

Money, Australian-style

Wherein Australian personal finance is surveyed in the manner of a practical ledger, and the legacy of a 19th‑century gold‑commodity economy is recorded as shaping tax, real estate and investment practice.

Neural denoising diffusion models with non-Gaussian distributions

Wherein neural denoising diffusion models are examined on non‑Euclidean manifolds, their applicability to partial differential equations and non‑Gaussian spaces is outlined, and a recent discovery by the author is noted.

Git tricks

Wherein a pragmatic workaround is described for keeping a repository’s .git metadata off cloud‑synced folders by using a bare repository plus a separate .git file, so normal git operations are preserved.

Pathwise solutions of stochastic differential equations

Wherein stochastic dynamics are treated pathwise by smoothing Brownian trajectories and solving the resulting ODEs, so that Wong–Zakai approximations are shown to converge to Stratonovich limits.

Scalable vector graphics

“SVG” to its friends

Wherein the SVG format is presented as a hub for vector interchange, and its interoperability with PDF and with conversion utilities such as Inkscape, dvisvgm, and web‑based generators is detailed.

Generative AI workflows and hacks 2025

Wherein practical notes are presented on installing local LLMs via Ollama and LM Studio, command-line pipelines using files-to-prompt → llm are detailed, and model-cost implications are recorded.

Text data processing

Wherein the peculiar arts of text data processing are surveyed, and command-line contrivances for CSV, JSON and LLM‑assisted scripting are evoked, including jq, xsv, llm and ttok.

Transformer networks

Baby’s first foundation model

Wherein the architecture is described as employing self‑attention and positional encodings, and scaling laws are invoked to permit extremely large, trainable language models.

Won’t somebody think of the children?

The bottom rung of the status ladder

Wherein the rhetorical elevation of children is examined, and the invocation “for the children” is shown to be used to justify policies from climate planning to AI regulation, irrespective of measured effect.

Multimodal AI

On the convergence between text, image, and audio and whatever else, as intelligence learns to get very general indeed

Wherein a shared semantic space between text, image, and audio is constructed by contrastive pretraining and is used to condition diffusion generation via CLIP embeddings, enabling cross‑modal retrieval and control.

Differentiable learning of collective automata

Wherein classical cellular automata are relaxed to continuous states and are trained by gradient-based learning using local Sobel-filtered gradients, and emergent patterns are framed as dynamical attractors.

Plotly

A pretty good plotting system

Wherein Plotly is presented as a plotting system whose browser‑based interactivity is described and whose non‑browser image export via orca or kaleido and multi‑language wrappers are noted

Designing social movements

Synthetic biology for utopian egregores

Wherein the mechanics of recruitment and institutional hygiene are examined, and Schelling-style coordination models are invoked to explain how movements self-replicate and police membership.

Email blogs and newsletters

Wherein email incarnations of blogs are surveyed, platform trade‑offs are catalogued, and Substack’s lack of an API, Ghost’s open‑source hosting option, and Buttondown’s privacy‑first defaults are reported.

Schelling-Goodhart coordination problems

Line in sand go up

Wherein coordination failures on shared belief maps are examined, and the use of Schelling points to render questions unaskable in culture wars and taboos is described, with attention to misrepresentation and preference cascades.

Science communication of ML research

PR for the robots

Wherein methods for relaying AI and ML findings to publics and peers are examined, with attention to safety and economics, and a noted tendency to slip into debate rather than sustained canvassing.

Indonesia

Wherein the emergence of online shopping in Indonesia is documented, difficulties in international shipping to Australia are noted, customary musyawarah mufakat practices are described, and past state violence is evoked.

Feedback system identification, not necessarily linear

Learning dynamics from data

Wherein nonlinear feedback systems are addressed via state filtering and variational methods, and recursive estimation of fixed parameters is shown to be achievable using simulation-based or RKHS-instrumental techniques

Melbourne / Naarm

Australia’s counterculture capital

Wherein the city’s dual naming as Naarm is noted and it is observed that the bush lies farther from town than in Sydney, so native superb wrens are encountered only rarely.

uv, the python package manager

A hipster python package manager with handy features for my workflow

Wherein the Python package manager is described, and its deep PyTorch GPU-specific support is noted; it is written in Rust, is claimed to replace many tools, and is presented as version-managing.

Neural denoising diffusion models with non-Gaussian distributions

Wherein diffusion models for categorical and Poisson-like data are treated, and physics‑inspired embeddings in N+D space with tunable D are described to connect Poisson flow and diffusion approaches

Privacy while web browsing

Browsing the internet without giving corporations my personal information for free

Wherein the browser is outfitted with extensions and compartmentalized containers to limit fingerprinting and third‑party trackers, and HTTPS‑only connections are advised for sensitive exchanges.

Human enthalpy

Nudges, changing things, UX for life

Wherein small amounts of activation energy are described as producing large shifts in behaviour, the region‑beta paradox is invoked to explain threshold effects, and personal anecdotes and policy examples are cited.

Machine learning for partial differential equations using diffusion models

Neural physics for neural physics

Wherein diffusion models are employed to learn PDEs, Poisson flow generative models are noted, manifold-valued solution spaces are considered for non-Euclidean PDE learning, and PFGM++ code is cited.

Voice transcriptions and speech recognition

Wherein speech is converted to text, real-time dictation and batch transcription are surveyed, with Whisper for recordings and Talon, Serenade and voice-programming workflows noted, and background music issues described.

Neural denoising diffusion models of language

Wherein a class of neural diffusion techniques is adapted to discrete text tokens, and a non‑Gaussian, discrete diffusion process is described as an alternative to autoregressive generation in low‑data regimes

Generative art with language+diffusion models

also some autoregressive models

Generative art using modern diffusion-backed image generators. The name-brand models are DALL-E 2, Stable Diffusion, Midjourney etc., which are diffusion models for image…

Free images

Raiding history for clip art

Wherein sources of freely usable images are catalogued, and the Rijksmuseum’s high‑resolution scans of 17th–18th‑century Dutch engravings are noted as accessible via translated search queries.

Neural nets that do symbolic mathematics, logic and other reasoning tasks

Wherein models are described as being taught to double-check their own reasoning by appending the token Wait at test time, and a curated s1K dataset plus budget-forcing are reported to boost competition-math performance.

Multi-agent self

Wherein the self is portrayed as divisible agents, and psychotherapy’s Internal Family Systems alongside shard theory are offered as compositional frameworks for explaining intrapersonal roles and conflicts.

Syncthing

A good, easy, peer-to-peer, free Dropbox-replacement system for the modern paranoid age

Wherein Syncthing is presented as a peer-to-peer file synchroniser without a central server, using TLS encryption, and is used to synchronise music-production folders across studio, backup and gig machines.

Python packaging, environment and dependency management

“import antigravity” raises an exception if you want CUDA antigravity

conda pip+venv poetry uv.

Pip-like python environment management

The mainline family of python package managers

pip is the default Python package installer. It is the most widely used and most widely supported package manager for Python, especially if you include the many derivative…

Multi agent causality

Game theory and decision theory for lots of interacting agents

Wherein agents' decisions are modeled via a mechanised multi-agent influence diagram (MMAID) extending causal DAGs to iterated games, and the problem of commitment races is examined for AI safety.

Neural PDE operator learning on domains with interesting geometry

Can ML solve PDEs on non-Euclidean domains?

Wherein implicit diffeomorphisms are learned to map Euclidean grids onto irregular manifolds, and neural operators are adapted to mask and represent partial differential equations on such domains.

Neural PDE operator learning

Especially forward operators. Image-to-image regression, where the images encode a physical process.

Wherein the Fourier transform is employed to learn PDE forward-propagators as resolution-agnostic operators, and the Burgers’ equation is used as a concrete example of one-step propagation learned from simulation timesteps.

The AI tech soap opera

Wherein the corporate choreography of AI is described, with Microsoft’s partnership with OpenAI and leaked executive correspondence recounted as elements shaping the industry’s unfolding narrative.

Feed readers

A standard for user-driven, open news

Wherein the revival of subscription feeds is described, Feedly’s ranking algorithm is noted as surfacing academic specialist blogs, and self‑hosted readers such as FreshRSS are outlined for private use.

Learning with conservation laws, invariances and symmetries

Wherein conservation laws are examined in the context of nonparametric and overparameterized neural models, and methods such as Lagrangian networks and Hamiltonian Monte Carlo are described

Causality, agency, decisions

Exotic decision theories, Newcomb’s boxes…

Wherein mechanised causal graphs are used to extend notions of causality to agents whose choices feed back into outcomes, and decision paradoxes such as Newcomb-style problems are examined.

Aligning AI systems

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

Wherein practical methods such as reinforcement learning from human feedback and mechanistic interpretability are surveyed, computational morality is noted, and the inherent fuzziness of alignment is acknowledged.

Conda-like Python environments

A parallel python package system to pip

Wherein the conda model is described, and it is noted that the distribution and the package manager are separately licensed, so commercial licensing risk for Anaconda distributions is made explicit.

Poetry, the python package manager

A python package manager that I briefly used

Wherein poetry is presented as a single-file project manager, is noted to create and manage local virtual environments, is described as offering its own dependency resolver, and is taxed by CUDA/GPU installs.

Leadership

The default pseudo-content for undirected MBA research

Wherein the efficacy of leaders is examined, and it is shown that monarchs' cognitive ability—instrumented by royal inbreeding—is linked to state performance in reigns where power is unchecked.

Reading ebooks

Wherein desktop handling of e‑books is presented, options are catalogued, Calibre’s ISBN‑oriented management is noted, and Thorium’s EPUB‑3 support and accessibility for LCP‑protected files are described.

Anti-TESCREALists

Post-rationalism for non-post-rationalists

Wherein the coinage TESCREAL is presented as an acronym for Transhumanism, Extropianism, Singularitarianism, Cosmism, Rationalism, Effective Altruism and Longtermism, and its othering function is described.

Computational complexity of Bayesian inference

Wherein the difficulty of Bayesian inference is examined, and the contrast between sampling-based posterior approximation and exact calculation on graphical models and on arbitrary measure spaces is presented.

Neural PDE operator learning using transformers

Wherein transformer-based operator learning for PDEs is examined, experimental integration with existing notebooks and benchmark comparisons are proposed, and diffusion-style alternative models are contemplated.

Adaptive design of experiments

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

Wherein adaptive experiment design is presented as surrogate-based optimisation for expensive noisy black-box functions, Gaussian process surrogates and acquisition functions are surveyed, and Ax is noted for deployment.

Causal Bayesian networks via probability trees

Wherein causal relations are expounded through sequential probability trees, interventions are enacted by fixing branches and temporal ordering is shown to afford identifiability in lieu of do-calculus.

Travel hacks

esp for technomads

Wherein practical travel hacks are catalogued, and techniques for compact shirt folding and procuring travel eSIMs are described, with notes on comparing Airbnb internet speeds using browser extensions.

Inference from disorder

Wherein inference is drawn from lack of structure, and the arrow of time is inferred from asymmetries in noise‑outsourcing conditions, while methods for causal ordering and entropy are examined.

Artificial agency

Wherein the question of agency is examined via causality-based models, the emergence of self in machines is contemplated, and the possibility that the human is not the agent in collaborations is considered.

Neural codecs and compression algorithms

Neural bandwidth reduction

Wherein neural networks are employed to compress images, audio, and video for transmission, a 90× audio codec is presented, and images are resampled into flexible-length 1D token sequences via vector embeddings

Machine learning for partial differential equations via flows

Wherein a normalising‑flow approach is applied to PDE learning in function space, and Gaussian process priors with Matérn kernels are employed so that function‑space ODEs map Gaussian noise to target solutions.

Causal inference on DAGs

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

Wherein graphs are invoked to decide when effects are identifiable from observational data via d-separation and do-calculus, and criteria such as back-door and front-door adjustment are described.

Scaling laws for very large neural nets

Theory of trading-off budgets for compute size and data

Wherein the behaviour of neural networks is examined as parameters approach billions, and an observational scaling method using about 100 public models is proposed, in which a low-dimensional capability space and sigmoidal compute trends are identified to predict novel capabilities

The simplest thing

Minimum viable whatever, worse is better, PC-losering, Postel chaos, Burkean engineering, maintenance externalities, DX vs UX

Wherein the problem of determining the least effortful course of action is examined, and the concrete test case of rewriting software is presented to show who is made to bear irreducible complexity.

Bayesian and causal inference by foundation models

Wherein transformers are examined as generalized inference machines and is argued that in‑context learning is modeled as Bayesian conditioning, with twisted sequential Monte Carlo methods proposed for sampling.

Data summarization

a.k.a Data distillation. sketching

Wherein methods for compressing datasets into weighted subsets such as coresets or representative subsets are surveyed, and connections to Bayes duality, sketching, and influence-based selection are outlined.

Extraversion

Also, extroversion

Wherein the nature of extraversion is recounted, its linkage to dopamine-driven reward learning is described, and the author’s personal stake as a high-scoring extravert is disclosed.

Causal inference under feedback

Wherein feedback loops are examined and it is shown that effective control can erase observational correlation, and extensions to continuous spatial fields and formal treatments by Central European causality researchers are surveyed.

Causal inference in highly parameterized ML

Wherein the application of causal graphs to neural nets and other nonparametric models is considered, dataset shift and invariance principles are examined, and benchmarking platforms for discovery are surveyed.

The Netherlands

The Dutch, the Low Countries

Wherein the origins of several capitalist practices are traced to the Low Countries, and the first major financial bubbles are observed, while colonial ties to Indonesia are noted and Rijksmuseum holdings are consulted.

Peter Piper’s Practical Python Pickling Patterns for Parallel Processing

Wherein Python pickling for parallel processes is described, cloudpickle is shown to serialize lambdas for distributed execution, and joblib plus compression methods are advised to reduce inter-process transfer overhead

Localhost dev server

My own little fake internet

Wherein local HTML pages are served for testing, and a simple python3 -m http.server bound to 127.0.0.1 is used, while considerations of TLS, Caddy auto‑SSL, tunnelling, and live‑reload are catalogued.

Unix commands I need often

but which are tedious to work anew out each time

Wherein modern replacements for classic Unix utilities such as ripgrep, fd and bat are catalogued, and techniques for file‑watching, resource monitoring and cross‑platform prompts are described.

Majority illusions and filter bubbles

Toxic cascades, squeaky wheels and social dark matter; the difficulties of estimating what all people think from the ones you see

Wherein the reader is apprised that, in homophilic social networks, a few highly connected hubs cause rare opinions to be made to appear common in local samples, thereby biasing prevalence estimates.

Email clients on linux

Unix invented email, and kinda left it there for the next 50 years

Wherein the available Linux email clients are surveyed, and the desirability of local offline clients with calendar and contact integration is contrasted with their prevalent usability and sync shortcomings.

Urbanism

Stuff about cities

Wherein urban order is surveyed, and debates from smart‑city surveillance to charter‑city experiments are presented, while city scaling laws linking population to energy use are examined.

Technological determinism

Technium, amistics, the robo-omega-point etc

Wherein technological change is presented as steered by tools already forged, and energy flows and urban-scaling data are adduced as concrete measures, while a personal notebook is reported lost.

Western Australia

A big state with modest dreams

Wherein the vast state is described; its landmass of 2,527,000 km² is noted and the capital, Perth, of about 2.14 million souls, is observed to favour secluded suburban life.

Collaborative intelligence

Between humans and machines

Wherein the partnership of human judgment and machine pattern‑finding is examined, and the rise of reinforcement learning from human feedback as a method of alignment is described, alongside risks of reverse‑centaurs.

Hydra for configuring and tracking machine learning experiments

Wherein Hydra is described as a generic configuration system for ML, a dedicated output directory per run is created to record the exact parameters and artifacts, and optional plugins for parallel parameter sweeps are supported.

Calendars and contacts GUIs for Linux

Wherein various Linux GUIs for calendars and contacts are surveyed, and CardDAV/CalDAV interoperability is reported, noting iCloud quirks and Flatpak-era GNOME/Evolution configuration pitfalls.

Gibbs posteriors

Bayes-like inference with losses instead of likelihoods

Wherein the posterior is constructed from a loss in place of a likelihood, and a tempering learning rate ω is introduced to control the influence of empirical risk, thereby linking inference to energy‑based models.

Organising a music collection

Also guessing missing metadata

Wherein music files are catalogued and analysed for key and BPM, associated artwork is managed en masse, and tools such as beets and MusicBrainz Picard are surveyed for metadata correction and bulk renaming.

Southeast Asia

Wherein the author’s long residence across Australia, Indonesia and Thailand is recorded, and a 17th‑century Swiss soldier’s itinerary in Dutch Formosa is noted as a historical curiosity, while the contention that Australia is excluded is registered.

Advice on advice

Wherein the perils of generic counsel are expounded, and a hierarchy of advisory skill — from reaction to causal modeling — is set forth, while the difficulty of personal calibration is noted.

Visual Studio Code for prose

for pros

Wherein VS Code is treated as a prose workbench, and its support for Markdown mathematics and a variety of spell‑checking extensions (cspell, SpellRight, LTEX) is catalogued for editing workflows.

Chromium browsers

Wherein the dominance of a single engine is noted, and alternative Chromium-based browsers are presented with concrete particulars such as Vivaldi’s built-in mail and calendar and Brave’s Linux emoji quirks.

DJing

On encouraging people to listen to your living room playlist by demanding that they pay for it

Wherein the craft of DJing is delineated as a global soundtrack of both capital and its opposition, and the practical note is given that Pioneer-compatible USB keys are carried to gigs.

Continuing to do obviously complicated things in a naïve way that doesn’t work

Also, not researching alternatives

Wherein a decade of sclerotic cohousing meetings is endured before governance literature is consulted, and the persistence of unresearched amateur strategies is described in measured, passive terms.

Gaussian Ensemble Belief Propagation

A method which improves our ability to solve important tricky problems, such as whether your house will flood tomorrow based on what the satellite says today

Wherein a novel Gaussian ensemble belief propagation is presented, its ensemble updates are used to update million‑pixel fluid‑simulation emulators, and it is noted that the work is accepted at ICLR 2025.

Fediverse

Mastodon et al

Wherein the fediverse is described as a volunteer-run, ActivityPub‑based federation of services such as Mastodon, Lemmy and PeerTube, and is hosted across myriad independent servers maintained by volunteers.

Python CLIs

Putting the “argument” in “command-line argument”

Wherein Python command-line interfaces are surveyed and their roles in configuring ML experiments with Hydra or pyrallis are noted, while multiple parser libraries and dependency overlaps are catalogued.

Misinformation, disinformation etc

Trolls, bots, lulz, infowars and other moods of the modern networked ape

Discussion of terrorism and hate speech

Attention management tips for web browsing

Wherein attention on the web is corralled through browser tweaks, extensions, and site blockers, simple tracking tools are described to record usage, and intrusive Google sign‑in popups are shown to be blockable to prevent modal interruptions.

Voice fakes

Wherein voice fakery is surveyed as a technical practice, and style-transfer methods are described as enabling cloning from seconds of speech but are noted to demand multi-gigabyte GPUs and tedious training.

Windows Subsystem for Linux

Wherein a miniature Linux distribution is reported to be installed and run within Windows via WSL, and the convenience of executing native Linux setup commands and tools from a Windows terminal is noted.

Distilling Neural Nets

Wherein a procedure is related in which a larger teacher model is employed to synthesize training examples, and a smaller student model is thereby instructed to emulate its probabilistic outputs for deployment.

submitit

A less horrible way to parallelize on HPC clusters

Wherein submitit is presented as a job-submission wrapper for SLURM, and the executor is employed to serialize and reload job handles to disk so queued Python tasks can be resumed from a later session.

Diffusion of innovations

Epidemiology of widgets

Wherein the geographic concentration of economically impactful technologies is recorded, the slow fifty-year dispersion of related jobs across regions is described, and an initial high-skill bias is observed to decline.

Operationalising the bitter lessons in compute and cleverness

Amortizing the cost of being smart

Wherein it is argued that massive training compute is justified because inference is cheaply amortised, enabling foundation models trained once to serve many tasks, and token-level economics and compute overhangs are examined

VS Code online and networked

Wherein remote usage of VS Code is treated, and a procedure for attaching a Python debugger via debugpy on a forwarded port is outlined, with SSH tunnels and tmux persistence being noted.

Making good intertemporal decisions

Description TBD

Wherein the intrapersonal collective‑action problem of conflicting selves across time is described, and interventions like role‑playing an advising counsellor and contractual self‑commitments are examined.

Sparse regression

Wherein penalised regression with LASSO-style absolute-coefficient penalties is presented, and sparse models are attained by tuning a penalty parameter under quadratic or generalised losses for predictor selection.

Regression with functional data

Wherein curves are regressed against predictors by being represented in functional bases, with derivatives and Karhunen–Loève components being used to capture dynamics and to permit warping-based alignment.