Mind Matters

Exploring the Dual Nature of Our Brain

Narrator

In the vast landscape of the human brain, two distinct parts exist: the limbic brain and the neocortex.

Mrs. Parker

The limbic brain is an ancient remnant, surviving with fewer neurons to create essential emotions like fear and happiness.

Nikki

But then came the neocortex, a newer, slower part, enabling us to plan and think through complex situations.

Narrator

So, while the limbic brain reacts swiftly, the neocortex takes its time to analyze and understand.

Mrs. Parker

This fascinating interplay means we often feel first and then figure out why we feel that way.

Nikki

But what if we could train our brains to harmonize these two parts more effectively?

Narrator

This is where wisdom emerges—training the limbic brain with the neocortex's insights.

Mrs. Parker

So, how does this relate to machine learning? Imagine a smaller model making quick, decisive actions.

Nikki

But this smaller model periodically validates against a larger one, known as Speculative Decoding.

Mrs. Parker

Imagine defining the essence of what it means to be human in a small core and giving it access the entire index of human knowledge!

Nikki

Wouldn't it be fabulous if the model can use tools and search for any information it wants to answer the query?

The brain consists of two parts. The limbic brain and the neocortex. The limbic brain is part of, an old Evolutionary process. It was able to survive wiith a Smaller set of neurons. Was able to do that by creating features like, Emotions, fear pain, Sadness. Happiness. Excitement.

Humans were able to stand up, use their hands. To control tools. And somehow. Improve a newer part of the brain called the new cortex. Which is slower, But is smarter. You can plan ahead. And, Think through what would happen with this new part of the brain. 

The lymbic part of the brain is faster and executes faster than the neocortex. And then essentially controls the neocortex.

We feel first. And then we figure out why. 

The limbic part of the brain, however, can be controlled by the neocortex as well, albeit slowly. Some of us manage figure out our own biases. And then train ourselves. To get around them.

That's where our wisdom comes from. We train the lymbic part of the brain. Using our neocortex. And we transform. From being Like any other Mammal, To humans that can leverage the neocortex to train the brain and get rid of biases.

Let's apply this to the field of machine learning.

Speculative decoding outputs Fast responses using a smaller model and then validates those using a larger model. If the model is accurate enough, it can go for some steps before getting validated by the bigger model.

There are processes to get the smaller model to update as well, using the responses from the bigger model. If the feedback loop is closed. The faster model is also is built by taking the smaller model and teaching it to iterate multiple times across the solution space before responding. And that iteration can happen within the model itself.

As time goes on, small models would evolve into a logic core. That can get just access concepts via search. Imagine defining the essence of what it means to be human in a small core and let it access the entire index of human knowledge via an index.

@Mrs. Parker @Nikki

Theme to add to all image prompts:  +Ben Shahn:Painters