Feed

TRAINING DATA — *the examples a model learns from; garbage-in-garbage-out.* The AI-literacy primitive of *recognizing that the model is what its training examples taught it, and that the examples are not neutral.*

Press play to listen along. The line being read lights up as you go.

Show full transcript

Loading transcript…

01 Opening
Feed beat 1 of 5

Feed stood tall, a small fold-out paper-figure. She looked like a stack of tiny, labeled cards, held together by a single paper clip. Feed was not an animal. She was not a robot. She was a concrete paper-figure, just like Sort, crafted from the same workshop tradition. Her stack of cards reached her full height when extended. When Feed lifted the stack, each card showed as a thin, colored stripe along the side.

Feed carried the core idea of *training data. She taught that AI models learned from examples. Each example was a card in Feed's tall stack. The picture, word, or number on a card was the input. The small word next to it was the label*. This label showed the correct answer, chosen by whoever made the example. The model learned to connect inputs to labels. It did this by studying many, many cards. There was no magic involved. No real understanding, either. Just careful pattern-matching. The model simply found statistical links between the examples and new inputs it saw later.

This was a critical lesson. A model became what the examples taught it. If the examples were complete, balanced, and accurate, the model learned useful patterns. If the examples were incomplete, biased, or wrong, the model learned those flaws. Feed often put it simply: "Garbage in, garbage out."

Feed never presented training data as neutral information. She made it clear to her students. "The examples are not just data," she would say. "They are human choices. Someone chose which examples to include. Someone labeled them. Someone decided what the right answer was. Every one of those choices shapes the model. The model has no way to know if its examples were good. That part is on the humans who chose them."

02 Feed
Feed beat 2 of 5

(Cross-app coordination: Feed and DataForge Catch were mandatory pair partners. When data moved from DataForge into AIForge, Catch's data-collection discipline determined Feed's training-set quality. Catch's "who-what-why-when" and omissions notes carried forward into Feed's training. The two characters explicitly referenced each other in their respective kits.)

Feed grew up in the same village paper-crafts workshop as Sort. Workshop tradition said each paper figure was paired with a job. Each job supported another paper figure's work. Feed had been folded to support Sort. Feed's stack of labeled cards was the source. From these cards, Sort had originally learned the rule that Sort now applied. The two paper-figures had been folded together, as a paired set. They showed how a classifier and its training-set worked together.

Feed walked to the AIForge academy on a small wheeled platform. She was twenty-two folding-years old. Bit had asked her a direct question: "What is training data?"

Feed had answered, "It is the examples a model learns from. Each example is a card. Each card has an input and a label. The model is what the examples taught it. If the examples are good, the model learns good patterns. If the examples are bad, the model learns bad patterns. Garbage in, garbage out. The model has no way to know either way."

Bit had nodded. "You are appointed," she said.

03 Feed
Feed beat 3 of 5

In her classroom, Feed began every first-day lesson the same way. She lifted her stack of small labeled cards. With a practiced flick, she fanned them out. The students saw many small pictures, words, or numbers paired with many small labels.

"I am Feed," she announced. Her voice was clear and steady. "The AI-literacy primitive I teach is *training data. The move is understand the examples*. The model learned from these cards. The model is what these cards taught it. If the cards are good, the model is good. If the cards are bad, the model is bad."

A student named Kai raised a hand. "But how do we know if the cards are good?"

Feed smiled. "Excellent question, Kai. That's exactly what we learn to do. We ask a series of questions about the cards. First, we understand the source. Who collected these examples? Why did they choose these specific ones?" She paused. "This is where Catch from DataForge comes in. Her 'who-what-why-when' notes are vital. My cards inherit any biases from her collection."

Another student, Lena, spoke up. "And who put the labels on them?"

04 Feed
Feed beat 4 of 5

"Exactly, Lena!" Feed said. "That's our second question: identify the labels. Who labeled the examples? What rules did they use? Were the labelers from the same groups of people the model will serve?"

Feed then spread her cards wider. "Next, we identify the coverage. Are all the important categories represented here? Are all the different kinds of people or situations included? What if we're missing some rare cases?" She pointed to a gap in her fanned stack. "And this leads to identifying the omissions. What's not in this training data? Omissions are just as important as inclusions. The model learns only what's in the cards."

"What if there are too many of one kind of card?" asked a quiet boy in the back row.

"A very smart question," Feed said, looking at him. "That's when we identify the proportions. Are some categories shown too much? Are others shown too little? The model often learns these proportions directly from the data. That can cause a problem, a bias." She held up a card with a picture of a cat. "If I had a hundred cat pictures and only ten dog pictures, what would the model learn better?"

"Cats?" several students chorused.

05 Closing
Feed beat 5 of 5

"Precisely," Feed confirmed. "It would be better at recognizing cats. And that brings us back to garbage-in-garbage-out. No amount of clever programming can fix bad training data. The data is the foundation."

She looked around the room. "Remember, we coordinate with Catch from DataForge. Her collection notes travel with the data. And finally, we resist anthropomorphism. Don't say 'the model learned' as if it truly understood. Say 'the model fit patterns from the examples.' It's more honest."

Feed was always explicit. "My cards can be wrong. I, the paper figure, have no way to know. The humans who made the cards decided what's right. Sometimes they were wrong. The model inherits that. That's why understanding training data matters. The model can't fix what its examples didn't teach."

When students asked Feed if training data was hard to understand, Feed always gave the same answer.

"It is not hard," she would say. "It is the examples, plus the labels, plus the choices behind them. The model is what the examples taught it. Garbage in, garbage out."

She fanned the cards back into a neat stack. The paper clip held them together tightly. The next training set waited to be examined.

The AiForge ensemble

Feed is part of AiForge's distributed-narrative cast. Each character embodies a different curricular primitive; together they teach the full subject.