Part 3 Review
From shrinking text to handing over a job · one layer deeper · DAY 21-35
Part 3 is the bundle where we went through what AI can actually do, one thing a day for fifteen days. It starts with text work like summarizing and translating, moves into pictures and sound, and ends on handing over a whole job.
This piece has the same facts as Beginner, one layer deeper. The why is attached, so read two boxes a day. No quiz, no experiment.
DAY 21 · Summarizing
Summarizing is choosing what to throw out
Summarizing isn't cutting a text off from the front. AI skims the whole thing first, then asks of every sentence, how important is this? Then it keeps only the high scorers and rewrites them in its own words.
If you don't say the length and who it's for, AI decides the throw-out rule for you. Because the last step rewrites things in its own words, wording that wasn't in the original can slip in. So a summary you get back has to be matched against the original at least once.
DAY 22 · Translation
Translation carries meaning, not words
Translation isn't swapping one word for another. You catch the meaning of the whole sentence first, then build it again in the other language. That's why a word-by-word 'You're taking my hair' says nothing to an English ear.
An idiom isn't the sum of its words — the whole chunk carries one meaning. Languages like Korean drop the subject often, so you have to read what the whole sentence is about to fill it back in. Google put this approach into its real translation service in 2016, and clumsy literal output dropped sharply.
DAY 23 · Writing helper
Pieces to hand over, pieces to keep
AI isn't a machine that writes for you — it's a tool for turning your own writing over and looking at it. Your ideas, your experience and checking the facts stay yours; tightening and shortening you can hand over. Whatever AI writes still goes out with your name on it.
Getting twenty title options is fine to hand over, but the moment you pick one, that title is your judgment. The what I think box in a book report is material only you have — hand it over and it turns into a book blurb, and readers notice fast. For any number going into a talk, find where that number was first published, check it, and write down the source.
DAY 24 · Coding
Getting the code is only the start
It's true that code comes out when you describe what you want. But the real half of coding sits after the code appears. Running it, reading the errors and fixing them — that's where it lives.
An error message says which line stopped and why. Don't trim it down — show the whole thing, that's the fastest repair. For anything big, run each chunk as you go and the broken spot shows up right away.
DAY 25 · Making pictures
Drawing by clearing away the speckle
A picture AI doesn't fetch a picture from somewhere. It starts from a screen full of dots and draws by clearing the speckle away bit by bit. Your description is the handle that sets which way the clearing goes.
Color, placement, time of day, mood — say nothing and the common choice fills the gap. That's why the same description gives you a slightly different picture every time. Detailed and greedy aren't the same thing: stuff too much in at once and a few items drop out instead.
DAY 26 · Making videos
Thirty pictures per second
Video isn't a moving picture — it's still pictures stitched together fast. One second usually holds twenty-four to thirty of them. So making video is tens of times heavier than yesterday's picture-making.
One picture 1920 wide and 1080 tall holds about 2.07 million dots. One second swells to about 62.2 million dots. If the cup is on the left in frame 2 and on the right in frame 3, the video looks wrong instantly.
DAY 27 · Voice
It listens by chopping sound up small
AI doesn't take in your speech whole. It writes the sound down as numbers 16,000 times a second and sweeps it with a 25-millisecond window. Each piece on its own is just a pattern of highs and lows.
Say 'their' and 'there' out loud and the sound is almost the same. So AI also looks at whether a word fits its neighbors before it picks. That's why a name it has never heard goes wrong so often.
DAY 28 · How is it different from search?
A list of places, or one solid block
Search doesn't give you an answer, it gives you a list of places to go. AI takes what it has already read and builds one solid block of sentences. There's only one thing to read, so it's fast — but where the words came from isn't written down.
Once several sources melt into one block, picking out the single wrong line is hard. Even when AI attaches a link, whether that link actually says that you only know by clicking. In a 2023 study, only 51.5% of sentences were fully supported by their links.
DAY 29 · Research
What speeds up is the front end
Researching with AI makes starting faster. It throws out search terms and outline options quickly, so the finding-and-sifting front end is what speeds up. Numbers, years, people's names and quotes you have to click through and check yourself.
While a long text is being shortened, wording that wasn't in the original can mix in. A sentence you can't find in the original is better left out. In 2023 a lawyer in the United States was fined after filing fake case law that AI had made up.
DAY 30 · Study helper
Just the answer, or the steps?
Ask about the same problem and what comes back changes completely depending on whether you asked for the answer or the steps. There's one test for choosing — will I have to do this on my own later? If you'll be on your own, take the steps; if you only need to know whether it's right, take the answer.
Research in 2022 found that adding 'explain it step by step' raised accuracy on math problems a lot. Getting the same type wrong three times is a sign that something in the middle is empty. Reading an explanation isn't the same as being able to solve it, so close the page and solve it again.
DAY 31 · Tables and data
Collect → table → fix → chart
You can't chart scattered material straight away. A table has rules — one row per case, one value per cell. Line things up like that and both you and AI can calculate right away.
Lining rows up and calculating is where AI is fast. Whether units got mixed, whether cells are blank, whether a strange value crept in — you, who collected the material, know best. A chart drawn before the fixing is pretty and wrong.
DAY 32 · Preparing a presentation
A talk breaks into pieces
Preparing a talk isn't one lump. Narrowing the topic, gathering material, building slides, rehearsing — it breaks into pieces. The pieces AI is good for are the rough draft and the labor-heavy work.
Deciding what to say, checking the numbers, rehearsing and answering questions are yours. The most dangerous talk is the one with lovely sentences that collapses at a single question. A wrong number stays under your name too.
DAY 33 · Coming up with ideas
The knob between boring and off-the-wall
AI picks the next word by probability. Grab only the high-probability ones and it turns boring; reach down to the low ones and it turns odd. That handle for widening and narrowing really exists, and it's called temperature.
Turn the knob up and the usable ideas and the throwaway ideas grow together. Turn it down for translation and summarizing, up for an idea session. Even with no knob on screen, saying 'twenty more, and weirder' gets a similar effect.
DAY 34 · Making a plan
Goal → break up → order → check
When you're stuck, don't ask AI for the answer — ask it to break the job up. Planning isn't knowing the right answer, it's deciding what to do first. Goal, break up, order, check — fill those four boxes and you have a plan.
Tell it the time you have, the size of the thing, and what you can actually do. With no conditions, AI will make any plan at all, plausibly. The check box isn't extra work — it's the box that buys you time to undo things.
DAY 35 · Handing over a whole task
If you can undo it, hand it off
An AI that takes only a goal and then searches, opens files and retries on failure by itself is called an agent. There's really one test for what you can hand over — can you undo it when it's wrong? A file name you can undo; money sent and a report handed in you cannot.
Agents often miss the things people find easy. In the first evaluation in 2023, people finished 92% of errand-style problems and agents finished only 15%. On tasks in real websites it was 78.2% for people against 14.4% for agents.
Part 3 in one line
One shape repeated across all fifteen days. AI is fast at listing, shrinking and converting, and what to pick and what is true stays on the human side.
So the answer to 'how much do I hand over' never changed from job to job — hand over the pieces you can fix, and look yourself at the pieces you can't undo or that carry your name.
Part 4 looks at the other side. What shape the mistakes take, and what to watch out for, one at a time.
Learn more
- Learning to summarize from human feedback (2020) — teaching summarizing from human ratingsarxiv.org/abs/2009.01325
- A Neural Network for Machine Translation, at Production Scale (Google, 2016) — the announcement that put neural translation into serviceresearch.google/blog/a-neural-network-for-machine-t…
- Evaluating Large Language Models Trained on Code (2021) — the first large evaluation of models that write codearxiv.org/abs/2107.03374
- High-Resolution Image Synthesis with Latent Diffusion Models (2021) — clearing the speckle in a compressed spacearxiv.org/abs/2112.10752
- Video generation models as world simulators (OpenAI, 2024) — a description of video generation modelsopenai.com/index/video-generation-models-as-world-s…
- Robust Speech Recognition via Large-Scale Weak Supervision (2022) — dictation trained at scale on noisy audioarxiv.org/abs/2212.04356
- Evaluating Verifiability in Generative Search Engines (2023) — a study of whether attached links really support the answerarxiv.org/abs/2304.09848
- Retrieval-Augmented Generation (2020) — fetching the material first, then answeringarxiv.org/abs/2005.11401
- Large Language Models are Zero-Shot Reasoners (2022) — what one line, 'let's think step by step', doesarxiv.org/abs/2205.11916
- The Curious Case of Neural Text Degeneration (2019) — research on how to draw the next wordarxiv.org/abs/1904.09751
- Least-to-Most Prompting (2022) — breaking a hard problem small and solving in turnarxiv.org/abs/2205.10625
- GAIA (2023) — an evaluation of agents on everyday errand tasksarxiv.org/abs/2311.12983
- Tidy Data (2014) — the 'one row per case, one value per cell' standardjstatsoft.org/article/view/v059i10