<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>From Bayes to Deep Networks :: Probability &amp; Probabilistic Computing Tutorial</title><link>https://josephausterweil.github.io/probintro/deep/index.html</link><description>The kiosk and the trainee. After the new year, two arrivals change campus life: a photo-recognizing bento kiosk in Cafeteria A, and a robot mascot trainee assigned to shadow Chibany. Neither can be reasoned with — and yet both clearly learn. This Part opens the black boxes: how a network sees, why its training is a likelihood in disguise, what attention actually attends to, and why a large language model doing in-context learning looks suspiciously like the hierarchical Bayes Chibany already knows.</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Thu, 16 Jul 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://josephausterweil.github.io/probintro/deep/index.xml" rel="self" type="application/rss+xml"/><item><title>From Bentos to Vectors: the Linear Algebra You Need</title><link>https://josephausterweil.github.io/probintro/deep/vectors-and-spaces/index.html</link><pubDate>Sat, 04 Jul 2026 00:00:00 +0000</pubDate><guid>https://josephausterweil.github.io/probintro/deep/vectors-and-spaces/index.html</guid><description>The Kiosk That Cannot See The new year brought two arrivals to campus. In Cafeteria A there is now a bento kiosk — a sleek box with a camera, said to recognize any bento you show it. And trailing three steps behind Chibany everywhere is the robot mascot trainee, assigned to learn the mascot trade by watching.
Chibany unwraps the lunchtime bento (the label says tonkatsu — students have labeled their bentos since the Markov chains chapter) and holds it up to the kiosk’s camera. The screen thinks for a moment, then chirps: tonkatsu. Chibany turns the box around, inspects the camera lens, peers behind the kiosk, and finds… no eyes. No taste buds. No nose.</description></item><item><title>From Rules to Weights</title><link>https://josephausterweil.github.io/probintro/deep/from-rules-to-weights/index.html</link><pubDate>Thu, 16 Jul 2026 00:00:00 +0000</pubDate><guid>https://josephausterweil.github.io/probintro/deep/from-rules-to-weights/index.html</guid><description>The Kiosk Nobody Programmed The bento kiosk in Cafeteria A has been running for a week now, and Chibany has developed a routine: hold up the bento, wait for the chirp — tonkatsu — and then check the label to confirm. The kiosk is almost never wrong. This morning, though, something new is bothering Chibany, and it takes a lap around the machine (trainee robot trailing three steps behind) to put a paw on it.</description></item><item><title>Neural Network Fundamentals</title><link>https://josephausterweil.github.io/probintro/deep/neural-net-fundamentals/index.html</link><pubDate>Thu, 16 Jul 2026 00:00:00 +0000</pubDate><guid>https://josephausterweil.github.io/probintro/deep/neural-net-fundamentals/index.html</guid><description>The Trainee Wants a Turn The previous chapter ended with the robot mascot trainee looking oddly eager, and this evening in Cafeteria A the reason comes out. Jamal has barely opened the laptop when the trainee rolls up, extends a manipulator, and taps the screen — right on the line of code that says jax.grad. Then it taps its own chest panel.
Jamal: “…You want to be trained. Not watch the kiosk get trained — be trained.”</description></item><item><title>Transformers &amp; Attention</title><link>https://josephausterweil.github.io/probintro/deep/transformers-attention/index.html</link><pubDate>Thu, 16 Jul 2026 00:00:00 +0000</pubDate><guid>https://josephausterweil.github.io/probintro/deep/transformers-attention/index.html</guid><description>The Journal Wants an Autocomplete The trainee robot has discovered the Bento Journal.
It happened quietly. Chibany was writing up the day’s entries — the trainee reading over their shoulder, as usual — and when Chibany’s pen reached “Friday, lunch:”, the trainee’s screen lit up with a suggestion: tonkatsu?
Chibany froze, pen in the air, then opened the furoshiki. Tonkatsu. The trainee had guessed the entry before the box was open — and over the following week its guesses kept landing about seven times in ten, exactly the campus base rate from Mystery Bentos. The trainee has invented autocomplete: predicting the next word of a sequence from the words so far. Chibany, naturally, wants a better one — one that could finish a whole Journal entry.</description></item><item><title>LLMs &amp; In-Context Learning</title><link>https://josephausterweil.github.io/probintro/deep/llms-in-context-learning/index.html</link><pubDate>Thu, 16 Jul 2026 00:00:00 +0000</pubDate><guid>https://josephausterweil.github.io/probintro/deep/llms-in-context-learning/index.html</guid><description>The Chatbot Learns a Word That Doesn’t Exist The bento kiosk can see. As of the New Year break, the campus can also talk: the university switched on a campus chatbot — a large language model, an LLM, wired into the student portal — and within a week everyone was asking it for cafeteria hours and club-room bookings. Chibany has been circling it the way they circled the kiosk: not suspicious, exactly, but unmeasured, and Chibany does not leave things unmeasured.</description></item><item><title>World Models &amp; Imagination</title><link>https://josephausterweil.github.io/probintro/deep/world-models-imagination/index.html</link><pubDate>Thu, 16 Jul 2026 00:00:00 +0000</pubDate><guid>https://josephausterweil.github.io/probintro/deep/world-models-imagination/index.html</guid><description>The Trainee Rehearses Lunch The robot mascot trainee has developed a strange new habit, and Chibany has been watching it for three days now. At the doorway of Cafeteria A, just before the lunch rush, the trainee stops. For half a second it goes still — no whir, no step — and then it walks in and threads the crowd to the bento kiosk without a single wrong turn. Chibany, who navigates the same crowd by bumping into people and apologizing, finds this deeply suspicious. What is the trainee doing in that half-second?</description></item></channel></rss>