Are We Algorithmic Beings?

 The team working at OpenAI is creating a human, and.....

I was learning about LLMs from none other than Andrej Karpathy, who co-founded OpenAI - he has uploaded a Deep Dive on LLMs like ChatGPT on Youtube.

The philosophy student in me can't help but notice drastic similarities between an LLM and a human.

The more I learn about that model, the more I learn about a human being.

These are some of the parallels I found:

1. Pre-Training

GPT - In its nascent stage the model ingests vast amounts of raw, unfiltered data.

Human - Absorbs fragmented sensory inputs as soon as he is born, without any filter.

Both systems inherit biases from their environments. A model trained in America will most likely know English better, and the first word of a human baby born in India will likely be in an Indian language. 

2. Post-Training: Supervised finetuning

GPT - The model is then taught how to converse, how to answer a specific question, how to behave, what questions not to answer etc.

Human - Similarly we are taught the "same things".

3. Hallucinations

GPT - LLMs invent false details when uncertain. (Very common in the earlier models)

Humans - Filling memory gaps, justifying decisions with invented explanations, seeing ghosts in the dark because that's where we are uncertain.

4. No Self Awareness

GPT - The model doesn’t know it is a model; it simply operates based on patterns.

Human - Humans also don’t truly know who they are—operating on assumptions of free will and identity without direct access to their "source code" (genetic/cultural programming).

5. Unaware of the creator

GPT - The model doesn't know who created it unless explicitly programmed with such information. 

(If you remember "Krutrim AI" made by Ola stated that it is made by OpenAI that can be because its initial data set be filled with text stating LLMs being made by OpenAI.)

Humans - Similarly we are not aware of our creator, just operating on another assumption of a creator with different names around the world - the name given by the people around.

If you learn more, you will find a lot of similarities - and you may feel that at some point differences would be rare.

We say that LLMs lack conscious experience, they merely simulate understanding without subjective awareness. These models are highly influenced by thier training data and biases within it. They are stochastic - predict the answers using probabilities and mathematical formulas and the output can be different each time.

Karpathy inadvertently revived the age-old debate of determinism vs. Free will.

What are we? 

What is the nature of our own "programming"? 

Are we also stochastic parrots?

Is there an irreducible spark that makes us unique?

As these models improve overtime, these questions will only become more important.