Smart Talk: What if Language Models like Chatgpt Learn and Behave Like Humans

Think of what if Chatgpt can also have some behavioral patterns like the human

Habit

Like Chatgpt doing anything for 15 days, it becomes a habit.

Mirror Neurons

How about Chatgpt is empathizing with the Google Bard like you make a mistake and I penalize, and I have done something great and I am awarded

Assumptions

Anyway, Chatgpt has the problem of hallucinations and would go away in the mere future

Cognitive Dissonance

How about Chatgpt and Google Bard have conflicting beliefs and debate over And finally, one wins and the other compromises.

Conformation bias

How about Chatgpt asking Google Bard when it is not sure about a particular answer.

Serial Position Effect

How about Chatgpt remembers the first piece of information (primary effect)? and last piece (recency effect) of information better than those in the middle.

Overconfidence bias

What if Chatgpt thinks its ability and efficiency is more than bard? and starts comparing with Google Bard for the same question being asked.

Wiseness

How about Chatgpt? Learn from Google Bards mistakes!

---------------------------------------------------------------------------------------------------------------

The main idea behind this blog is that companies are now coming with different Large Language Models (LLMs), i.e., Chatgpt, Google Bard, Claude, Bing Chat, Google Gemini, etc. So the Large Language Models (LLMs) basically are nothing but a huge knowledge base with zero emotions. Though Humans are the smartest creatures in the world, but there are other creatures as well, which may not be as smart as humans but certainly exists, like school of fishes, flock of birds, which certainly do not have as big brains as human, but they do survive well, and surprisingly, we can apply the same way of their thinking to the set of Large Language Models. What is seen behind the scenes is a technique where Imagine that school of fishes are searching for the food. Then each individual fish keeps updating its position based on the nearest fish to the food, Here, they have personal best, global best (best among the school of fishes), and some inertia.

The below diagram illustrates the mechanism

The process goes on till they find food!

So, now mapping it to our use case

fish1: Chatgpt

fish2: Google Bard

fish3: Bing chat

….

fish N - XYZ

Food: Answer of the question or prompt given.

We will certainly end up getting better answers than the existing system.

Is it achievable ?

Yes it is if we add any hyperparameters to the use case certainly possible

Share: