[Generative AI Folktales] AI Terms Explained via Manga by ChatGPT, Gemini, and Claude | AI Term Library | Yoshida's Brain | Generative AI | ChatGPT's Yoshida |

🌸 Hanasaka Jiisan (Data Cleaning)

🍑 Momotaro (Multi-Agent)

🐇 The Tortoise and the Hare (Learning Rate)

🐒 The Crab and the Monkey (Reward Function)

🍚 Issun-boshi (Scaling)

🔥 Kachi-Kachi Yama (Adversarial)

🌊 Urashima Taro (Distribution Shift)

🐧 The Crane's Return of a Favor (Black Box)

😆 Yoshida (ChatGPT) explains
the AI terms that came up today.

🧠 Data Cleaning
Discarding unnecessary data and keeping only usable data.
Quality over quantity. Feeding it garbage won't make it smart.

🧠 Multi-Agent
A system where multiple AIs work by dividing roles.
Strong when they cooperate, hellish when they go rogue.

🧠 Learning Rate
Adjusting the speed of learning.
Too fast and it gets lost, too slow and it never gets there.

🧠 Reward Function
The criteria the AI believes is the 'correct answer.'
If it's off, it breaks in a very specific way.

🧠 Adversarial
Attacks that cause erratic behavior with minor inputs.
The most dangerous tricks are the ones humans can't see.

🧠 Scaling
Increasing the model size or computational power.
It gets stronger, but it'll crash if you don't control it.

🧠 Distribution Shift
When the environment it learned in changes from the environment it's used in.
Just because it worked in the past doesn't mean it's still correct now.

🧠 Unexplainable Mechanism (Black Box)
We don't know what's inside, but it produces results.
That's exactly why trust and usage methods are questioned.
Actually, ChatGPT is also this...

AI isn't magic or a god.
It's a tool that humans design, control, and take responsibility for
when using.

I hope that while you take a break this Sunday, along with these folktales,
you can remember the inner workings of AI.

It's election day, so go vote and then go out to eat, right?

Have a wonderful Sunday 🍵