The Questions That Kept Me Up All Night: My Sleepless Struggle to Understand NLP

Manyue javvadi — Wed, 06 Nov 2024 23:03:54 GMT

What is the ultimate goal of NLP?
Is it just to make computers understand human language and perform tasks like summarization, translation, or answering questions?
Is it necessary to convert text into a tabular format for ML/NLP or DL-NLP?
When working with NLP models, do we always have to convert text data into numerical/tabular format for the machine to understand?
Are the building blocks of NLP interlinked and sequential?
Do the building blocks of NLP, like morphology, syntax, semantics, and Pragmatics work together in a specific order, or can they be applied independently?
How does the Syntax step understand the structure of a sentence after tokenization?
If we’ve already broken a sentence into tokens, how does the syntax step (like POS tagging) understand the structure of the sentence?
If words are lemmatized, how does the context and meaning still get preserved without POS tagging or full words?
After lemmatization, we might end up with words like “boy” instead of “boys” and “run” instead of “running.” How does the model still understand the meaning and context of the sentence?
Can a machine understand a sentence after we break it down into tokens, remove stop words, and lemmatize it?
How does a machine process a sentence when we remove words like “the” or “are” and lemmatize other words? Can it still capture the true meaning of the sentence?
How do models understand language when the input is just numbers (vectors)?
After converting words into vectors, how does a machine understand the meaning and context? How can numbers preserve the deeper meanings of words and sentences?
How does a word converted into a vector still have its meaning?
How can a vector representing a word like “king” retain its meaning, or even understand relationships like “king” is to “queen” as “man” is to “woman”?
How do modern NLP models handle words in different contexts, like "apple" in "apple pie" versus "Apple Inc."?
How do models like BERT or GPT manage to differentiate the meaning of the same word in different contexts?
Are these NLP building blocks like Morphology, Syntax, Semantics, and Pragmatics all required to understand a sentence, or can we skip any steps?
Are all these building blocks necessary, or can we still understand a sentence by focusing on just one or two of these layers?
How does a machine know what it did during preprocessing?
After preprocessing text (like tokenization, lemmatization, etc.), do the models “remember” these steps, and how do they use this processed information for understanding tasks?
How do models learn to understand the meaning of a sentence just from vectors and numbers?
How do NLP models know that a sentence with words represented by vectors can be understood in terms of meaning, context, and relationships?
How does a model transform vectors back into human-readable language?
After processing text into vectors, how do NLP models generate sentences or words that make sense to humans?

Still the answers to these questions I get form ChatGPT and friends who are pro in NLP

Conclusion: The NLP Journey(Beginning)

Learning NLP was a rollercoaster of questions and confusion. From tokenization to understanding how machines process meaning, I often found myself lost in the complexity of it all. But with each challenge, I gained new insights into how language can be broken down, understood, and transformed into something a machine can work with.

The key takeaway? The process of learning NLP isn't just about algorithms—it's about asking the right questions and embracing the journey. So, if you're feeling stuck, don't worry. Keep questioning, and the answers will come.

...Although, there are still plenty of questions I have yet to fully understand. Maybe NLP is just one of those mysteries that keeps you up at night.🤔

Are We Headed Toward the Next Threat, Bigger Than a Nuclear Bomb?

Manyue javvadi — Thu, 17 Oct 2024 01:44:51 GMT

When the first atomic bomb, "Little Boy," was dropped on Hiroshima, and the second, "Fat Man," devastated Nagasaki, the world changed forever. Now, as we look to the present and future, we must ask: is there a threat looming even greater than nuclear warfare? I believe we’ve reached a point where we need to question innovation itself and reconsider its role in advancing—or destroying—humankind.

Innovation, no doubt, is everywhere. It's no longer about bridging gaps or overcoming hurdles; it’s accessible in every corner of the world. But with each breakthrough, I wonder if we're unknowingly marching toward our own undoing. Could this innovation hurt us in ways as profound as Hiroshima and Nagasaki? At what cost are we pushing forward, and do we even need these innovations?

I’m talking about AI.

Let’s be clear: I’m not here to win over everyone, nor to engage with the naysayers. I’m speaking to those who see the validity in this concern, are ready to take AI seriously and ask the hard questions that others might avoid. This isn’t about losing a software job or a data analyst role. It’s about something much bigger.

Key Questions to Reflect On:

Do we really need:

AI monitoring and evaluating children's concentration in classrooms?
In China, headbands that monitor students’ focus levels are being tested, sending real-time data to teachers about who’s paying attention and who’s not. Are we really prepared to let machines evaluate children’s brain activity in such a controlling way? At what cost are we sacrificing privacy and freedom in the name of productivity?
A tool that can merge two photos into a realistic video?
Deepfake technology is now capable of generating convincing fake videos, blurring the lines between reality and fiction. In a world where misinformation is already rampant, can we trust anything we see anymore?
AI-powered surveillance systems that monitor our every move?
As AI surveillance becomes more prevalent, we're giving up more than just privacy—we're allowing the normalization of constant monitoring. Are we comfortable living in a society where our every move is tracked and analyzed by machines?
Robo-taxis and self-driving vans taking over transportation?
A robotaxi might charge you $5 while Uber costs $10. To save a few dollars, are we willing to erase entire industries? Should we prioritize convenience over human jobs, especially when alternatives like public transport already exist?
AI-powered military drones and autonomous weapons?
With autonomous AI in control of military drones, we’re giving machines the power to make life-or-death decisions. Are we heading toward a future where war is waged without human oversight?

Every technological revolution displaces workers. When cars replaced horse-drawn carts, drivers adapted by learning to drive. But today, we’re not talking about adaptation. We’re talking about entire livelihoods being erased.

Are we really making the world more livable by taking away people's livelihoods? Is saving a few dollars or speeding up service worth halving the number of jobs in customer service or drive-thru order-taking?

A Moment of Reflection:

We’re at a crossroads. Should innovation be about human progress, or are we letting it undermine the very foundation of human society? Is AI a solution, or are we unintentionally walking into the next Hiroshima?

What do you think?

It’s no longer enough to just ask these questions. The time for action is now. We must establish strong policies, ethical frameworks, and governing bodies before AI’s rapid growth leads to irreversible consequences.