Phase 2 of MT24 Learning is complete: Foundations of Knowledge.

Bayes and base rates, models that are wrong but useful, the replication test, and Popper, Kuhn, Lakatos, Feyerabend.

Sprints 4–7 of 90.

#MT24Learning #PhilosophyOfScience #Epistemology

Sprint 5/90: What Is a Model?

I compared the textbook pendulum model with the exact period in six lines of Python. Under 1% off below 20°. 18% off at 90°.

Wrong everywhere, useful in a known range.

#MT24Learning #PhilosophyOfScience #Physics

Sprint 6/90: The Scientific Method, Objectivity & Its Failures

2015: 100 psychology studies rerun. 97% of originals were significant, 36% of replications.

My own simulation: test 20 null effects, get a "finding" 64% of the time.

#MT24Learning #OpenScience #Reproducibility

Sprint 11/90: axioms and declared assumptions.

Euclid's parallel postulate resisted proof for two thousand years. Beltrami's 1868 model showed why: a geometry without it is as consistent as Euclid's own. Change that one axiom and triangles change.

#MT24Learning #Mathematics #Geometry

Sprint 8/90 · Calibrating intuition.

I scored three simulated forecasters on 10,000 events. Always 50%: 0.250. Honest: 0.188. Overconfident: 0.203. The loud one said 96% and was right 84%.

Before I automate a judgement, I score my own.

#MT24Learning #Forecasting #Probability

Sprint 7/90: Four Minds on Science

Popper, Kuhn, Lakatos, Feyerabend.

I checked Einstein's 43″ per century for Mercury in six lines of Python: 42.97. The number that killed the planet Vulcan.

#MT24Learning #PhilosophyOfScience #HistoryOfScience

Phase 3 summary · Calibrating Judgement.

98% ranges that miss 40% of the time. Gut feelings that deserve trust only with fast, clear feedback. Formulas that beat experts on average, and still need calibrating.

Next: Phase 4, maths and logic.

#MT24Learning #Forecasting #CriticalThinking

Sprint 9/90: linear algebra and calculus for ML.

In a narrow loss valley, gradient descent zig-zags: each step points downhill, not at the minimum. I computed the path and drew it.

And: backprop only computes the gradient. Descent does the learning.

#MT24Learning #MachineLearning #LinearAlgebra

Sprint 10/90: logic, propositions and inference.

Wason's four-card task: E, K, 4, 7. The rule: vowel on one side, even number on the other. The most common pick is E and 4. The right pair is E and 7, because only they can break the rule.

#MT24Learning #Logic #CriticalThinking

Sprint 4/90: Knowledge, Truth & Evidence

Five lines of Python: 1% base rate, 90% sensitivity, 5% false positives. One positive test gives 15.4%.

Gettier (1963) showed a justified true belief can still be luck.

#MT24Learning #Epistemology #PhilosophyOfScience

Sprint 1/90: environment + Git foundations.

What stuck with me: Git stores snapshots, not diffs, and every object is identified by its hash (SHA-1 by default, SHA-256 optional).

Next up: hashing in depth and GitHub over SSH.

#MT24Learning #Git #LearningInPublic

Sprint 3/90 closes Phase 1: project structure and quality gates.

The src/ layout makes tests run against the installed package, not loose files in the project root. Then four gates: ruff format, ruff check, mypy, pytest, repeated in CI.

#MT24Learning #Python #CodeQuality

Phase 1 done: 3 of 90 sprints.

Git stores snapshots named by hashes. SHA-1 is still the default, SHA-256 is opt-in. Only the .pub key goes to GitHub. And a check that hooks can skip, CI must repeat.

Next: Knowledge, Truth and Evidence.

#MT24Learning #Git #Python

Sprint 2/90: hashing + GitHub over SSH.

One changed letter gives a completely different SHA-256 hash. Git relies on this to name every object by its content.

Key lesson from the SSH part: only the public key ever leaves your machine.

#MT24Learning #Git #Cryptography