Post Image

AI is coming in hyper-exponential speeds.

OpenAI has announced GPT-4, a large multimodal model that can process text, images, audio and video data. GPT-4 is the latest and most advanced version of the generative pre-trained transformer (GPT) family of models, which have been widely used for natural language processing tasks.

GPT-4 claims to have achieved the best-ever results on capabilities and alignment, meaning that it can perform a wide range of tasks across domains and align with human values and preferences. Some of the tasks that GPT-4 can do include generating realistic images from text descriptions, synthesizing speech from text or images, answering questions from multiple sources of information, and creating interactive games and simulations.

GPT-4 training was completed last August. Today it is released for all of us. It is safe to say that internally they are already on GPT-6.

We must begin to realise the concept of exponential growth, as its something that we will talk about a lots in the near future.

Here’s another example of exponential growth with paper folding:

Suppose you have a piece of paper 0.1 mm thick. If you fold it in half once, it will be 0.2mm thick. If you fold it in half twice, it will be 0.4mm thick. If you fold it in half three times, it will be 0.8mm thick.

You can see that every time you fold the paper in half, the thickness doubles. This means that the thickness follows an exponential function of the form y = ab^x , where y is the thickness, x is the number of folds, a is the initial value (0,1) and b is the growth factor.

Now suppose you want to know how many times you need to fold the paper to make it as thick as Mount Everest, which is about 8848 meters tall.

The answer will surprise you

To answer this question, you can use the exponential function and solve for x:

y = ab^x

8848000 = 0.1 * 2^x

88480000 = 2^x

log(88480000) = log(2^x)

log(88480000) = x * log(2)

x = log(88480000) / log(2)

x ≈ 26.5

So you need to fold the paper about 27 times to make it as thick as Mount Everest.

This shows how quickly the quantity can increase due to exponential growth.

Now imagine that GPT3 had 175 billion parameters and GPT4 has 170 trillion. So we are talking about hyper-exponential growth.

This is how fast the new era will come in our life.

svgMaybe there are things even Chomsky should not talk about (AI).
svg
svgOpenAI CEO Admits to Being 'a Little Scared' of AI: CTO warns for a point of no return.
Georgios at the cover of MIT Sloan magazine

My book