Vector Playground – When AI shows how it understands

I believe in well-shaped projects,
in the elegance of code,
and in the joy of doing things your own way.

From there came Korail, the first no‑code CMS with native vector search.
An MTV‑ready project (multi‑language, transversal, vectorial), launching soon.

And working on Korail pushed me further: to test, go off-track, and make visible what usually remains hidden in the deeper layers of AI.

Each exploration in my Set Code & Sail Lab isn't just a game: it's a concrete way to show how AIs think.

With my latest module — Vector Playground — you can enter two or three words (for instance: "ocean", "fish", "computer") and see how their vectors get closer… or drift apart.

A new exploration: Vector Playground

---

Vector Playground — preview

With this module, you can enter two or three words (for example: "ocean", "fish", "computer") and see how their vectors get closer… or drift apart.

How it works?

• AI doesn't "see" words like we do. It turns them into vectors: sequences of numbers that represent meaning in a mathematical space.
• Two nearby vectors = two nearby ideas.
• Two distant vectors = two different worlds.

Cosine similarity (0 → 1)

• 1 = both texts mean the same thing
• 0.8 = they often appear in related contexts
• 0.5 = they have almost nothing in common

👉 In the module, you see two things: the raw cosine (between -1 and 1) and its normalized version (between 0 and 1), easier to compare.

---

Two models, two views of the world

We use two AI models, with model‑specific thresholds for badges:

ADA (1536D)

• ≥ 0.94 → very close
• 0.87–0.939 → close
• 0.72–0.869 → related
• < 0.72 → far

3‑Large (3072D)

• ≥ 0.89 → very close
• 0.77–0.889 → close
• 0.67–0.769 → related
• < 0.67 → far

Two different universes: ADA is fast and "generous" (it easily sees similarities), 3‑Large is stricter (it only groups what's truly related). Seeing both helps you understand how two AIs perceive the same concepts.

---

Live examples

In Vector Playground, try:

• ocean ↔ fish → close, makes sense.
• ocean ↔ computer → surprising… but AI has seen millions of texts where computers simulate oceans.
• banana ↔ computer → much farther apart (phew).

This is where exploration gets interesting: it's not a school test, it's a dive into how an AI associates ideas.

---

Automatic explanation

I added a function that makes the experience even clearer: an "Explain with AI" button.

After comparing three words, you get not only the cosine similarity scores, but also a plain‑language interpretation.

Example:

• ocean ↔ sailboat (0.825) → Close: sailboats often travel on oceans.
• ocean ↔ computer (0.821) → Odd but logical: computers model oceans in oceanographic research.
• sailboat ↔ computer (0.823) → Odd but logical: software helps plan sailboat navigation.

Result: the AI explains why it positioned these words this way in vector space.

---

Why it matters

If you grasp this, you already understand a big part of what powers:

• next‑gen search engines,
• chatbots,
• and how AIs generate text, images, or code.

They don't reason with words, but with clouds of numbers.

---

Try it yourself

I wanted to make it visual, playful, and free. You can add your own words to the dataset, run a search, and see how many attempts you have left for the day (so my server doesn't explode!).

---

Conclusion

Vector Playground is a small exploration.
But it gives a strikingly clear sense of how AIs understand us.

And if a fish, a sailboat, and a computer can coexist in a space of numbers… imagine what your content can become once it's translated into vectors.

---