Part 5 Review
From one dot to the phone in your pocket · short, with analogies · DAY 46-55
AIMade by AIThis part review was written by AI (Claude) from that part's lessons and records, and published after automated checks. The owner doesn't read every item.How it's made
Part 5 spent ten days on what the inside of an AI looks like. It started with what one dot does and ended up inside your own phone.
No quizzes and no experiments here. It's a read-and-move-on piece. If a day feels fuzzy, jump straight to that day's lesson.
DAY 46 · Neural networks
One dot can only multiply and add
Open up an AI and you find dots passing numbers around, stacked in layers. That stack is called a neural network. All one dot does is multiply, add, and check a threshold — that's the whole job. The smartness isn't in any single dot. It builds up as numbers pass through layer after layer.
DAY 47 · Attention
Attention is a share you split up
Attention is the math that decides how hard one word in a sentence stares at the other words. The staring doesn't stretch forever — the words split up a fixed share of it. Stare hard at one word and the rest get weaker, and where the staring piles up becomes the meaning.
DAY 48 · Meaning as numbers
Meaning gets turned into a place
A computer can't read letters, only numbers. So each word gets stored as a bundle of numbers. Words used in similar company learn similar numbers, so words with close meanings land in close places. Once meaning is a place, near and far become something you can calculate.
DAY 49 · Finding similar things
It measures meaning, not letters
Whole sentences can become number bundles too. Then you can measure the distance between two of them. Even with no letters in common, same meaning means a close place. And when the letters are nearly identical but the meaning flips, they move apart - like "that was fun" and "that was no fun".
DAY 50 · Answering with the book open
Look it up and read it, instead of reciting
Answering from memory alone makes it easy to invent something plausible when you don't know. So before answering, it goes and finds the relevant text first and lays it open beside the question. That turns a closed-book test into an open-book one. And it can say where the words came from.
DAY 51 · Teaching it extra
Nudge a grown model instead of raising a new one
You can take a base model that already did the big round of learning and teach it one more job. That's called fine-tuning. It isn't learning from scratch — only a tiny slice of the numbers inside gets changed. It's strong at changing tone and format; for new facts you use answering with the book open instead.
DAY 52 · People handing out scores
Not the right answer, just which one is better
You get two answers to the same question and have a person pick the better one, tens of thousands of times over. What you collect isn't right-and-wrong, it's what people prefer. Then you build a small scoring model that copies that preference, and tune the AI toward its scores.
DAY 53 · GPUs
The shape of the math was the same all along
A graphics card is a part built to do easy math a huge number of times at once. AI training has exactly that shape, which is how it became the heart — one card holds about 80 billion tiny switches. Even that wasn't enough: Llama 3 was built on 16,000 cards tied into one block, running for 54 days.
DAY 54 · Data centers and electricity
One is tiny, the pile is not
One text question came in at a median of 0.24 watt-hours - about a TV left on for 9 seconds. The catch is that questions like that keep piling up, nonstop. Data centers worldwide used 415 terawatt-hours of electricity in 2024, about 1.5% of all electricity.
DAY 55 · AI inside my phone
It splits the work into pieces
Today's phones carry a small model of about 3 billion numbers, so they can summarize and polish even with no internet. Long text or hard questions get sent out to that building from before and the answer comes back. Whatever it can finish inside, it finishes inside - faster, less electricity, and your words never leave.
Part 5 in one line
Squeeze ten days into one line and it's this - there was no magic inside. Just multiplying and adding, piled up an enormous number of times.
Part 6 looks ahead. Is an AI that matches people really coming, and why do grown-ups argue about it?
Learn more
- Deep learning (Nature, 2015) — a short summary written by the three people who built deep learningnature.com/articles/nature14539
- Learning representations by back-propagating errors (1986) — the paper that made fixing weights by the size of the error widely knownnature.com/articles/323533a0
- McCulloch & Pitts, A logical calculus... (1943) — the first paper to treat a nerve cell as a computing elementlink.springer.com/article/10.1007/BF02478259
- Neural Machine Translation by Jointly Learning to Align and Translate (2014) — the translation study where attention was first usedarxiv.org/abs/1409.0473
- Attention Is All You Need (2017) — the paper that dropped reading in order and kept only attentionarxiv.org/abs/1706.03762
- Efficient Estimation of Word Representations (2013) — the study that spread the idea of turning words into number bundlesarxiv.org/abs/1301.3781
- Sentence-BERT (2019) — a way to build sentence bundles that are good for measuring distancearxiv.org/abs/1908.10084
- Dense Passage Retrieval (2020) — finding text by places of meaning instead of letter matchingarxiv.org/abs/2004.04906
- Retrieval-Augmented Generation (2020) — the study that had a model answer with found text laid open beside itarxiv.org/abs/2005.11401
- LoRA (2021) — the method that cut the numbers you touch down to one ten-thousandtharxiv.org/abs/2106.09685
- Training language models to follow instructions with human feedback (2022) — tuning a model with records of what people pickedarxiv.org/abs/2203.02155
- NVIDIA Hopper Architecture In-Depth (2022) — the makers' own walkthrough of the card used for trainingdeveloper.nvidia.com/blog/nvidia-hopper-architectur…
- The Llama 3 Herd of Models (2024) — the record of a 54-day run on 16,000 cardsarxiv.org/abs/2407.21783
- Measuring the environmental impact of AI inference (Google, 2025) — a company-reported measurement of the electricity one question takescloud.google.com/blog/products/infrastructure/measu…
- Energy and AI - Executive summary (IEA, 2025) — an official summary putting data center electricity in national termsiea.org/reports/energy-and-ai/executive-summary
- Introducing Apple's On-Device and Server Foundation Models (2024) — an official post on splitting work between a phone model and a server modelmachinelearning.apple.com/research/introducing-appl…