With the increased hype over ChatGPT and AI projects that seem to be on a massive increase right now; the media hysteria over science fiction through the eyes of The Terminator, The Matrix and pretty much any dystopian AI fiction out there, and there are those willing to embrace it. So what is it anyway?
What is AI?
AI or Artificial Intelligence is a group of studies aimed to mimick the intelligence or behavioural patterns of humans and other living beings. There are different methods used to understand how we think and apply that to computers to help solve our problems.
What is Machine Learning?
Machine Learning is a subset of AI that focuses on using computers to learn from data sources without needing to use complex sets of rules. This approach is mainly based on training a model from sets of data.
What is Deep Learning?
Deep Learning is a technique that performs Machine Learning inspired by the our own brain’s network of neurons.
What is a Model?
In mathematics, there is a subset called statistics that relies on assumption. Assumptions are used to calculate probabilities. A simple example is the outcome of the throw of two 6 sided dice.
The assumption is that the probability of one dice landing on a face (1,2,3,4,5 or 6) is 1 out of 6, or 1/6. We can calculate the probability of both dice coming up with the same number as 1/6 x 1/6 or 1/36.
We can then calculate the probability of any event (2 and 3), or (1 and 6) etc.
The statistical assumption constitutes a statistical model, as we know there will be an outcome of rolling the dice.
The statistical model has been interpreted in two different directions for Deep Learning, the first is Feed Forward Neural Networks, where inputs and outputs are independent of one another; and Recurrent Neural Networks where the output is dependent on the prior elements of the given sequence. Recurrent Neural Networks are referred to as “memory”.
What is GPT?
GPT or Generative Pre-Trained Transformer. A transformer is a type of model that was first introduced by Google in 2017 in a research paper. 5 years later it formed the basis of ChatGPT.
Transformers are a Deep Learning Model that solves the problem of transforming input sequences into output sequences. The transformer breaks down the sequence of input into individual stages called vectors, then pays “attention” to those vectors to see how they interact with each other. The result is then decoded into the output sequence.
The first tool that used Transformers was BERT (Bidirectional Encoder Reproductions from Transformers) and then OpenAI introduced GPT-3 to the world.
Pre-Trained Models are Models that have been trained on large sets of data for a particular task.
What is ChatGPT?
ChatGPT is a chat bot that provides access to the GPT interface. Chat bots have been around for years as a method of artificial communication for humans but the first time with a Transformer Model. OpenAI also provided a method of accessing ChatGPT for programmers to use in their applications.
One of the problems with ChatGPT is that since it is based on assumptions, if there is something that it doesn’t have the correct answer to, it will use assumptions to create a plausible answer to the problem, however the answer will not be the correct answer. GPT-4 is now connected to the Internet, so the answers will be compared with search results to gain a better answer.
Thoughts about ChatGPT
ChatGPT uses something called NLP or Natuial Language Processing, which is just the way to write things in a certain way that is natural for humans to understand. You communicate through a language of “prompts” to steer ChatGPT in the direction you want.
In the early days of ChatGPT it was thought that you’d have to give ChatGPT examples of what you wanted to get a better result, and researchers would give a list of similar questions and answers to get ChatGPT to return a more accurate result. It turns out that ChatGPT can also be used with zero training to give a much better answer when the prompt mentions the output should be given by a master in that skill, or the result simplified.
How are Transformers used?
While ChatGPT relies on LLMs (Large Language Models) there are other use cases, with audio and visual language models. Already there are AI artists using Models to create new art and even photo realistic images. There are other uses in language translation and creating realistic videos in different languages.
Image recognition and video recognition allows AI to work better in self-driving vehicles, which can also be applied to self flying and self swimming vehicles.
What about AutoGPT?
AutoGPT is a task oriented tool since ChatGPT is a chat bot. AutoGPT solves tasks or achieves set goals using specific and structured answers.
Conclusion
The discovery of Transformers have enabled ChatGPT and others to become available to the general public. Just be aware that the results generated are based on assumptions and not reality.
AI may not live up to the marketing hype nor the Science Fiction dystopia that is often portayed in the media, so there needs to be a middle ground. It will change people’s lives, that is for certain.
Consider that people have been manipulated through social media before the advent of AI, I can see AI as being a useful tool when used correctly.
