LLMs playing text adventures, weather forecasts, making a base template in Yesod
Curious how the latest crop of LLMs do on text adventures, I set them at it. They're better, but not great. You'll also learn how to interpret meteograms, and how to set a base template for a Yesod site.
Entropic Thoughts Premium Newsletter #8
For this premium newsletter, I'm trying something new: I'll keep it more informal and conversational. Both to take some pressure off myself, and to see if...
Bind-to-LambdaCase, text adventures, and insurance claim frequencies
Everyone loves Haskell idioms, right? But we also love text adventures. And do you know how often major claims are made against home insurance? Soon you do!
Hazards, accidents, and data processing
We look at some fundamentals of system safety – everyone's favourite overlooked topic. Then we talk information theory because that's always a hoot.
Emacs blogging, ViewPattern argument transform, and named pipes
In case you didn't know, org-babel is amazing. How do we avoid naming intermediary values in code, which we don't care about? And what happens when we open a named pipe for writing?
Getting open data out of ECMWF, drunk driving, and mangled English
We dig into a classic faulty machine translation, we learn how to get some data out of ECMWF, and then drop an unexpected (?) statistic about drunkenness in traffic.
Procedural rhetoric and the accounting equation
We learn what procedural rhetoric is and how it affects the players of games. We also look at how to interpret the expanded accounting equation.
Technology warts are a good thing, actually
We learn why we should cherish the warts of our technologies, and what the difference is between Hong Kong and decimal odds. Plus new premium newsletter!
Entropic Thoughts Premium Newsletter #7
A games-heavy premium newsletter, where we look into MCTS for card-shedding games, teaching children poker, house rules to make Risk more complicated, but! also an approach to teaching children computers.
Nix package debugging, getting correlation from sources of variation
We figure out how to debug a Nix package build. But more excitingly, we also learn how we can measure variation with and without an effect present and get an intuitive sense for what the correlation means!
Search index in Haskell, estimating distances with milliradians
We build a search index in 150 lines of Haskell (not much!) and use clever properties of circles to estimate distances. Also you should subscribe to the church of the radian.
Conditional for pattern, a guessing game tutorial, and the reaction phase
We learn yet another Haskell idiom: the conditional for. We move on to implementing a guessing game in Haskell to understand what is meant by "most code is impure anyway". Finally, we get some wisdom from Ward about the lead time of product development.
Making Haskell hard to read, and information technology
Do idioms make code harder to read? Is information technology plug and play? Betteridge's law gives you the answer: no and no.
Evaluating LLMs on text adventures, what overproduction does to waiting
You get to watch me spend money on API credits as we learn which LLM models can play text adventures. We also get a striking insight into why overproduction is such a insidious waste.
Keyboard review, LLMs play text adventures, and how to do what you don't know
If you're looking for a good virtual Qwerty keyboard for Android, I've got you covered. If you want to know how well LLMs play text adventures, likewise. If you want to know why Toyota focuses on learning, well, read on!
Entropic Thoughts Premium Newsletter #6
Hello, premium subscribers! I have been spending some of my holiday time away from technology, so this premium newsletter concerns itself perhaps more than...
We drive text adventures by Perl script, and we infer Poisson rates backwards
In order to have an LLM play text adventures, we need a way to programmatically drive them. We figure it out! Then we learn how to infer a Poisson rate through just the probability of no event. It's useful, sometimes.
Homeowner's investment investigations and dangerous statistical comparisons
We look at the alternative strategies of paying off the mortgage in full first, or investing all of one's savings and paying off the mortgage in full 20 years later. We also learn what to call the dangerous power of statistics where it lets us draw conclusions without understanding anything.
On creating text adventures and predicting the past with the future
We look at what it takes to create a well-polished text adventure (and get a chance to play mine!) and we get brief but possibly useful advice from Bueno de Mesquita.
Financial calculations in Emacs; fundamental questions of information theory
We learn how to compute with compounding interest in Emacs (and discover that the HP-12c is still better.) The two fundamental questions of information theory also get a mention.
Drones, Hill plots, and optimising the wrong thing
We speculate about the cause of the Chinese push toward better drones. We look at a bunch of Hill plots. We take a hint from Optimising Oracle Performance.
Mario colours and engine-off procedures
We do a nerdy dive into Mario title colours and marry them with entropy. Then we rehearse the first step when an engine quits in a single-engine aircraft. In case our children pull the throttle when we're on the simulator, or something.
Sigh... monty hall. And functional programming!
We learn an intuition for the Monty Hall problem that is actually intuitive, in contrast to all other intuitions. At least I think so. We also look at why functional programmers get by with less formal verification.
Stock market index value and lean development
We look at the risk-adjusted returns of stock investments and how they compare to other instruments. We also learn a neat way to phrase the purpose of lean development.
Insurance and the deadly effect of speed
We learn how to make insurance decisions and we get our yearly safe driving tip, where we look at the mathematical shape of the number of deadly accidents as a function of increased road speed.
Org parsing, shorthand generation, and paired comparisons
I get annoyed by a breakage in Org which turns out to be reasonable. We look at a swoosh generated by JavaScript. And we learn how paired comparisons get their power.
Haskell and SQLite for the win, starring corecursion
We learn how to make a side project low maintenance with clever use of boring technology. We also get an aimless look at how to achieve corecursion with foldr.
Entropic Thoughts Premium Newsletter #5
We get a hint of my latest obsession (text adventures), learn a thing or two about Haskell and the EA movement, and get a review of Code That Fits In Your Head, a book with a good premise but which falls a little flat.
Group decision-making and writing time
We get a free, easy-to-use tool to decide on things anonymously in groups. We learn a little about the time required to read and write a page of technical text.
Mobile-friendly sidenotes, and distance estimation with milliradians
We learn how to make mobile-friendly sidenotes without requiring JavaScript (though requiring some HTML templating.) We also learn to estimate distances with no external tools but our hands and a little mental arithmetic!
Dealing with Haskell errors, and how 3G works so well
We learn to read Haskell errors by example, and then we find out which sequence of coin tosses is most likely – and what that has to do with 3G cellular communication.
The greatest procedural language in the world, and some obfuscation of it
We learn in what sense Haskell is the greatest procedural language in the world, and we debate whether or not to abstract over functions.
Survival, TLS certificates, and non-obvious interpretations
Two new articles: one on statistics and one more sysadminny. Then we learn what formulations are in ethnography.
Parser combinators vs. regexes, and slurping in Perl
We compare regexes to parser combinators for an early Advent of Code problem, and we learn some Perl peculiarities that help with reading files.
Good correlations and bad flashcards
This week's article is a long one, on correlations! Even if you think you know correlations, there's probably a thing or two you don't.
Insurance perils and US Army deployment planning
Domain jargon can be funny, as is the case with insurance perils. We also learn about how the US Army plans deployments – or didn't, in that one case in 2003.
Entropic Thoughts Premium Newsletter #4
Hello, premium subscribers! Aside from the link collection, I will not be recommending a book or two today, but tell you about my top 20 (twenty!) technical...
Emacs gushing, and secondary product development wastes
We take another look at how to use Emacs more, and we learn about the things that slow down software product development projects, despite hard work.
Web-based forecasting tool and model building tips
We get some tips on model building (specifically variable selection) and then also get our hands on a very sloppily made forecasting web tool.
Firewall configuration embarrassment and higher education
We learn how to avoid embarrassing ourselves when misconfiguring e.g. firewalls, and what higher education is for. (Among other things.)
Plus and minus takes the prize
We learn some of the most important research there is – on the superiority of linear models to expert judgment. We also learn how to easily find 30 % more problems when reviewing someone's code.
Deploying low-maintenance web apps and fluid dynamics
We lament how annoying it is to maintain side projects and celebrate by creating a side project that requires maintenance – although less than ordinarily. We also lament our lack of CFD knowledge.
Brier scores, language rants, and data-intensive applications
We learn about how we can compare forecasting accuracies, and a little about protocol design for avoiding forward compatibility.
Entropic Thoughts Premium Newsletter #3
Some links, a couple of book recommendations, a preview of a future article, and ACX 2025 forecasting contest rationales.
Experimental design for non-compliance, and Perl arcania
We learn a little experimental design for when subjects opt out of the treatment branch we have planned for them. We also learn about an extremely un-useful Perl feature.
Social constructions and high rates of scientific false positives
We take a brief look at how something seemingly-technical like "selling out" an event is a social construction, and we reason from basic principles how much of published science might be false positives.
Memory limits, the north pole is moving, and standard deviations
We learn a systemd trick and some history around the perception of physics. We also think about whether rote memorisation of standard deviations is a useful way to get a sense of variation, and if this in turn leads to better statistical literacy.
Statistical literacy and saving lives for $1,000,000
We hypothesise about something called statistical literacy, which may or may not be a developmental milestone. We also learn what it takes to save a life.
Entropic Thoughts Premium Newsletter #2
We get three links to good articles, we learn about talking empathetically, handling children, what reinsurance is, and get some forecasting tips.
The Basean Loom, and arguing for self-review
We learn how we can combine conditional probabilities into an implied base rate. Separately, we learn that self-review is about as efficient as peer review, although less effective.