News for hackers. Stuff that matters.
Re:Grok 4 Launch [video] by boroboro4 on Thursday July 10, 2025 @01:57PM (#44523704)
The fact the word ends up being 1 token doesn’t mean model can’t track individual characters in it. The model transforms token into vector (of multiple thousands dimensionality), and I’m pretty sure there are dimensions corresponding to things like “if 1st character an ‘a’, 1st is ‘b’, 2nd is ‘a’ etc.

So tokens aren’t as important.

Re:Grok 4 Launch [video] by kadushka on Thursday July 10, 2025 @04:10PM (#44524978)
Is there any evidence to support your hypothesis?
Re:Grok 4 Launch [video] by boroboro4 on Friday July 11, 2025 @08:42AM (#44531465)
Good question! I did a small experiment: trained a small logistics regression from embedding vectors into 1st/2nd/3rd character in token: https://chatgpt.com/share/6871061a-7948-8007-ab53-5b0b697e90...

I got 0.863 (for 1st)/0.559 (for 2nd)/0.447 (for 3rd) accuracy for Qwen 3 8B model embeddings. Note the code is hacky and might be wrong in ways + in reality transformers do know more because here I utilize only embedding layer. However it does show there are very clear signals on characters in tokens in embedding vectors.

Re:Grok 4 Launch [video] by kadushka on Friday July 11, 2025 @02:10PM (#44535292)
Thank you! I guess if there's enough spelling related text in the dataset, a model is forced to learn some info about token composition in order to predict such texts.

I wonder if it would help to explicitly insert this info into an embedding vector, similar to how we encode word position info. For example, allocate the first 20 vector elements to represent ASCII codes of token's characters (in some normalized way).

Re:Grok 4 Launch [video] by boroboro4 on Friday July 11, 2025 @02:29PM (#44535497)
One way here is to use one hot encoding in first (token length * alphabet length) dimensions.

But to be frank I don’t think it’s really needed, I bet everything really needed model learns by itself. If I had time I would’ve tried it though :)

Bonus content, accuracies for other models (notice DeepSeek!):

- Qwen3-32B: 0.873 / 0.585 / 0.467

- Qwen3-235B-A22B: 0.857 / 0.607 / 0.502

- DeepSeek-V3: 0.869 / 0.738 / 0.624

Re:Grok 4 Launch [video] by boroboro4 on Friday July 11, 2025 @04:55PM (#44536708)
Ok, bonus content #2.

I took Qwen3 1.7B model and did the same but rather then using embedding vector I used vector after 1st/etc layer, below accuracies for 1st positions:

- embeddings: 0.855

- 1st: 0.913

- 2nd: 0.870

- 3rd: 0.671

- 16th: 0.676

- 20th: 0.683

And now mega bonus content: the same but with prefix "count letters in ":

- 1st: 0.922

- 2nd: 0.924

- 3rd: 0.920

- 16th: 0.877

- 20th: 0.895

And for 2nd letter:

- embeddings: 0.686

- 1st: 0.679

- 2nd: 0.682

- 3rd: 0.674

- 16th: 0.572

Re:Grok 4 Launch [video] by brookst on Thursday July 10, 2025 @05:13PM (#44525683)
No, the vector is in a semantic embedding space. That's the magic.

So "the sky is blue" converts to the tokens [1820, 13180, 374, 6437]

And "le ciel est bleu" converts to the tokens [273, 12088, 301, 1826, 12704, 84]

Then the embeddings vectors created from these are very similar, despite the letters having very little in common.

Re:Grok 4 Launch [video] by boroboro4 on Friday July 11, 2025 @08:47AM (#44531526)
Character on 1st/2nd/3rd place is part of semantic space in generic meaning of the word. I ran experiments which seemingly ~support my hypothesis below.