Cheaper LLM labelling
We learn how to optimise classification by slotting in a fast classifier in front of the slow one, and continue our flashcard theme of evolution.
Tag index for Org mode blog
We learn a little more about how the Entropic Thoughts HTML is generated, and how Darwin made inferences about the distant past.
Better AI code comment detector
You get a free, privacy-oriented web tool to detect whether code comments are written by an LLM or not! And a great way to phrase what a confounder is.
Claude comment detection
We learn about a statistical model that separates Claude-generated code comments from human-written ones, and how much traffic is too much traffic when flow starts to suffer.
A curmudgeon tries a language server
Let's learn how to get closer to Lisp-style development in Haskell, and when to allow module A to depend strongly on module B.
Snakes and ladders
We improve snakes and ladders to become an actual game, and learn what to do with uncertainties in an unfinished specification.
Entropic Thoughts Premium Newsletter #14
Hello! Less substantive premium newsletter this time, clocking in at a bit over 4000 words. (My provider warned me that it was too long, but I managed to cut...
A noob learns FFT
We learn how to transform signals from time domain to frequency domain, along with pretty graphs and hopefully a tiny bit of intuition. And we (gasp!) take averages of averages.
Purely functional digital circuit simulator
We finally make the circuit simulator pure using a generator-consuming-a-lazy-sequence approach!
Digital circuit simulator in Haskell
Travel break. SICP-in-Haskell!
Kuiper Q-Q plot: are these the same?
Whenever we ask, "Is this thing, which I think is special, really different from this normal thing?" we are performing a generic hypothesis test. Here's a tool to do it easily! Also: brief, interesting history on tulipmania.
Intuition for distribution differences
Let's look at and compare cumulative distribution functions, and think about in which direction the divergence operator measures flux.
A full body MRI earns you a year of smoking
Exposing healthy people to full-body MRIs to screen for cancers might seem excessive, but it reduces risk about as much as not smoking for a year.
You could have invented type classes!
We look at how the authors of SICP almost invented type classes and some tricks for using them in Haskell. We also find out why the Apollo capsule position in lunar orbit was not known to greater precision than 10 metres.
Tagged data in Haskell (SICP 2.4.2)
We read a little part of SICP, the wizard book, and apply it to Haskell. Then we talk system robustness and liveness probes.
Entropic Thoughts Premium Newsletter #13-2
A sprawling newsletter covering computers, children, health, and a game that allows itself to be narratively complex.
Entropic Thoughts Premium Newsletter #13-1
A sprawling newsletter covering computers, children, health, and a game that allows itself to be narratively complex.
Lean, not backpressure (plus what congestion is)
We look at what Lucas Costa does to rein in code-generating robots, and learn where we can learn more about similar techniques.
LLMs write almost good code
I present a brief case study of when the LLM made things more complicated than necessary, and then we practice our intuition for risk sizing in betting.
More keyboard help! And Formula 1 winners from qualifying
I ask you again for help in completing the keyboard survey, we look at how to determine the effect of qualifying on winning a Formula 1 Grand Prix, and learn how much of a Fortran program is assignment.