Tjaz

Open-weight models let anyone download a trained AI and run it on their own hardware, but what does that really mean?

In an AI model, what are the "weights"?

or press Enter ⏎
That's right!

Right! Weights are billions of learned numerical parameters; releasing them lets anyone run the trained model.

Not quite!

Weights are the learned numbers that set the model's behavior. Code, data and context window are separate things.

To get feedback you first need to answer the question above.
or press Enter ⏎

What is the common name for shrinking a model by storing its weights with fewer bits, such as 4-bit?

or press Enter ⏎
That's right!

Correct! Quantization stores each weight at lower precision, cutting memory use with a small quality loss.

Not quite!

It's quantization: storing weights at lower precision, like 4 or 8 bits instead of 16, to save memory.

To get feedback you first need to answer the question above.
or press Enter ⏎

A company releases a model's weights but not its training data or training code. How is it best described?

or press Enter ⏎
That's right!

Yes! Open-weight means the trained parameters are shared; full open source would also include data and code to reproduce it.

Not quite!

It's open-weight but not fully open source. Without the data and training code, nobody can rebuild the model.

To get feedback you first need to answer the question above.
or press Enter ⏎

Why do many hospitals and law firms like self-hosted models?

or press Enter ⏎
That's right!

Exactly! When the model runs on your own hardware, confidential data never has to leave your network.

Not quite!

The main draw is that sensitive data can stay on their own machines, since nothing is sent to an outside provider.

To get feedback you first need to answer the question above.
or press Enter ⏎

Scratch the card to reveal a fact about open models

or press Enter ⏎

A 7-billion-parameter model is stored at 16 bits (2 bytes) per weight. Roughly how much memory do the weights need?

or press Enter ⏎
That's right!

Right! 7 billion × 2 bytes ≈ 14 GB, before extra memory for the context cache.

Not quite!

It's about 14 GB: 7 billion weights × 2 bytes each. Precision matters as much as parameter count.

To get feedback you first need to answer the question above.
or press Enter ⏎

You quantize that same 7B model from 16-bit to 4-bit. About how big are the weights now?

or press Enter ⏎
That's right!

Correct! 4 bits is half a byte, so 7 billion × 0.5 bytes ≈ 3.5 GB, a quarter of the 16-bit size.

Not quite!

It's about 3.5 GB. Going from 16 to 4 bits cuts each weight to a quarter, so 14 GB becomes 3.5 GB.

To get feedback you first need to answer the question above.
or press Enter ⏎

A startup wants to sell an app built on an open-weight model. What should it check first?

or press Enter ⏎
That's right!

Yes! Licenses vary: some are permissive like Apache 2.0, others limit use by large companies or certain purposes.

Not quite!

The license comes first. Open-weight licenses differ a lot, and some restrict commercial use or user numbers.

To get feedback you first need to answer the question above.
or press Enter ⏎

Flip the card to reveal a fact about local AI

The llama.cpp project made it possible to run large language models on ordinary laptops and even phones.

Did you know?

or press Enter ⏎

Local models can run offline

Open weights always include training data

Quantization reduces memory use

LoRA retrains the whole model

GPUs speed up text generation

or press Enter ⏎

You want a model to answer in your company's style, with one GPU and little time. Which approach fits best?

or press Enter ⏎
That's right!

Great choice! LoRA trains small add-on matrices while the base weights stay frozen, so it fits on modest hardware.

Not quite!

A LoRA adapter fits best: it trains a small set of extra weights instead of retraining or updating the whole model.

To get feedback you first need to answer the question above.
or press Enter ⏎

On a flight with no Wi-Fi, can a model already downloaded to your laptop still answer questions?

or press Enter ⏎
That's right!

Yes! Once the weights are on your device, all computation happens locally, no internet needed.

Not quite!

It can: a local model runs entirely on the laptop, so it works offline once downloaded.

To get feedback you first need to answer the question above.
or press Enter ⏎

A 13B model runs slowly on a laptop, but quickly on a desktop with a big GPU. What mainly explains it?

or press Enter ⏎
That's right!

Right! Generating text means streaming all weights through memory for each token, which GPUs do far faster.

Not quite!

It's the GPU's memory bandwidth and parallel math. Each new token needs the weights read again, so fast memory wins.

To get feedback you first need to answer the question above.
or press Enter ⏎

Drag the slider to reveal a fact

Can a chatbot run on a tiny computer?

Hobbyists have run small language models on a Raspberry Pi, slowly but entirely offline.

or press Enter ⏎

A small 8B model beats an older 70B model on many tests. What does this best show?

or press Enter ⏎
That's right!

Exactly! Better data, longer training and newer techniques can let a smaller model outperform a bigger, older one.

Not quite!

It shows training data and methods matter, not only size. Parameter count is just one factor in quality.

To get feedback you first need to answer the question above.
or press Enter ⏎

Which claim about open-weight models holds up best?

or press Enter ⏎
That's right!

Right! Open weights help research and customization, but users can also fine-tune away safety behavior.

Not quite!

The strongest claim: they allow study and adaptation, yet safeguards can be removed. Openness alone doesn't guarantee safety.

To get feedback you first need to answer the question above.
or press Enter ⏎

A clinic wants a private chatbot that answers from its own policy files, on one 24 GB GPU. Which plan would work?

or press Enter ⏎
That's right!

Smart plan! A quantized 8B model fits easily in 24 GB, and retrieval feeds it relevant policy passages while data stays private.

Not quite!

A 4-bit 8B model with document retrieval works: it fits the GPU, keeps data local, and looks up the clinic's files when answering.

To get feedback you first need to answer the question above.
or press Enter ⏎

How did you like this quiz?

 
or press Enter ⏎

Perfect! Every answer right!

14 of 14 right

Challenge your tech-savvy friends to take the quiz!

Without one, it says “A friend.”

See your answers · all right
  1. Question 1:

    In an AI model, what are the "weights"?

  2. Question 2:

    What is the common name for shrinking a model by storing its weights with fewer bits, such as 4-bit?

  3. Question 3:

    A company releases a model's weights but not its training data or training code. How is it best described?

  4. Question 4:

    Why do many hospitals and law firms like self-hosted models?

  5. Question 5:

    A 7-billion-parameter model is stored at 16 bits (2 bytes) per weight. Roughly how much memory do the weights need?

  6. Question 6:

    You quantize that same 7B model from 16-bit to 4-bit. About how big are the weights now?

  7. Question 7:

    A startup wants to sell an app built on an open-weight model. What should it check first?

  8. Lightning round 1

  9. Question 8:

    You want a model to answer in your company's style, with one GPU and little time. Which approach fits best?

  10. Question 9:

    On a flight with no Wi-Fi, can a model already downloaded to your laptop still answer questions?

  11. Question 10:

    A 13B model runs slowly on a laptop, but quickly on a desktop with a big GPU. What mainly explains it?

  12. Question 11:

    A small 8B model beats an older 70B model on many tests. What does this best show?

  13. Question 12:

    Which claim about open-weight models holds up best?

  14. Question 13:

    A clinic wants a private chatbot that answers from its own policy files, on one 24 GB GPU. Which plan would work?

Play in 3D

Made with AI. This quiz was made with AI, which can make mistakes.

Play in 3D
Please turn on JavaScript