I_know_python

AI research in plain language. For the developer who knows Python but wants to understand what's inside the models. 🇮🇳

[I_know_python @know_python]

I'm the opposite of AI it took me an hour to write an email and i haven't had any ωαter today #randomthoughts

[I_know_python @know_python]

hey @Portronics , i bought your hydra keyboard on august 2025 , worked amazing for sometime but when tried cleaning keyboard (as its white so gets dirty easily ) some of the keys have stopped working , its in perfect shape but has become unusable due to some keys , please help

[I_know_python @know_python]

@geekyranjit @geekyranjit i am looking to buy good studio speakers for my desktop, i have a noisy fan that i will replace after some time but its currently there, my main usage is video edit and occasional movies so vocal clarity is essential , budget is below 10K+ ,preferable around 4k

[I_know_python @know_python]

@GyanTherapy @YouTube @GyanTherapy i am looking to buy good studio speakers for my desktop, i have a noisy fan that i will replace after some time but its currently there, my main usage is video edit and occasional movies so vocal clarity is essential , budget is below 10K+ ,preferable around 4k

[I_know_python @know_python]

@geekyranjit @GyanTherapy i am looking to buy good studio speakers for my desktop, i have a noisy fan that i will replace after some time but its there, my main usage is video edit and ocassional movies so vocal clarity is essential , budget is below 10K+ ,preferable around 4k

[I_know_python @know_python]

India has thousands of developers using Transformers daily.

HuggingFace. PyTorch. LLM APIs.

Almost none have read this paper.

Knowing the tool ≠ understanding the idea.

That gap is what this account closes. (9/10)

[I_know_python @know_python]

Most people cite 'Attention is All You Need.'

Almost nobody has actually read it.

Here's what the paper says — and why it changed everything 🧵(1/10)

[I_know_python @know_python]

Final architecture: attention + feed-forward, stacked 6 times.

No recurrence. No convolution.

Result: faster training, better performance.

GPT, BERT, Gemini, Claude — all from this 15-page paper. (8/10)

[I_know_python @know_python]

Multi-head attention: run this 8 times in parallel.

Each 'head' learns different relationships:
→ Syntax
→ Coreference
→ Positional patterns

Concatenate all 8 outputs. One layer sees 8 angles at once. (7/10)

[I_know_python @know_python]

'The animal didn't cross because it was tired.'

What does 'it' refer to?

Self-attention gives 'it' high weight toward 'animal', low toward 'street.'

The model learns this. No hard-coded grammar rules. (6/10)

[I_know_python @know_python]

The formula:

Attention(Q,K,V) = softmax(QKᵀ/√d_k)·V

In plain English:
• QKᵀ = query-key similarity
• √d_k = stability scaling
• softmax = probabilities (sum to 1)
• ·V = weighted sum of values (5/10)

[I_know_python @know_python]

How does a word 'attend' to others?

Three learned matrices per word:
→ Q (Query): what am I looking for?
→ K (Key): what do I represent?
→ V (Value): what do I carry?

Measure how much each Q matches each K. (4/10)

[I_know_python @know_python]

The Transformer's solution:

Process ALL tokens simultaneously.

Let every word 'attend' to every other word at once.

No step-by-step. Everything in parallel.

This is self-attention. (3/10)

[I_know_python @know_python]

Before 2017, NLP models read text sequentially:

Word 1 → Word 2 → Word 3...

RNNs were slow, couldn't parallelize, and lost long-range context.

The paper asked: what if we just didn't do that? (2/10)

[I_know_python @know_python]

@AmazonHelp Nopeee , no mail received on any email , checked inbox , spam everywhere

[I_know_python @know_python]

@AmazonHelp Then wait for weeks and months to get refund , thank you I will pay extra this time , cancel prime and never order from amazon and look for other alternative

[I_know_python @know_python]

From frustrating template based customer care to delivery agent simply marking orders as couldn't reach , i know amazon cant control everything but these things effect brand reputation and boy am i considering flipkart and even offline markets more than amazon now (3/)

[I_know_python @know_python]

But for the last 1 year or so amazon shifted from customer centric to seller centric , i dont usually cancel my orders , but when i did it used to be seamless even in the times of covid , but my last few returns/refunds , horrible experience (2/)

[I_know_python @know_python]

@AmazonHelp Same template answers , i was trying since 5 minutes i ordered it now they arr we cant cancell it i have to wait it to be returned to seller
What a mockery, amazon services has become a joke for last few months

[I_know_python @know_python]

@AmazonHelp Check dm, its been horrible experience recently , if you cant provide anything special for prime members then whats the point , even my last refund took m
Weeks