Continuing the series on what is AI?

If you have read the first post in this series, you will probably tell me that you are already using some part of AI in your business. You definitely use a mapping service like Google Maps. You use Whatsapp.
Many of your IT systems are already AI enabled.

You possibly think of AI as being ChatGPT or Gemini and others and that you might not be using. Let us now look at a ChatGPT type AI system.
These AI systems use LLMs – this stands for “large language model”.

GPT stands for “Generative Pre-trained Transformer”, the aim of which is to respond with a human-like text. We have already discussed “Hey Google, what’s the weather like tomorrow?” and received an answer “It may rain, take a raincoat if you go out”. It gives what seems to be a human reply.

Of course it is not human. It has been programmed with a lot more than the IF ELSEIF ENDIF statements that we discussed previously. It uses a large language model, i.e. it has been given access to huge amount of data, or text, of phrases, sentences and paragraphs. So when you ask a question, it tries to match this with all the data it have access to, similar to, but a lot more complex than the Google search engine. Its responses seem to have a human quality because it sometimes uses the words “I” or “I apologise” or “Enjoy”. It also gets many questions and answers completely correct. It has to make use of fast processing, fast retrieval, fast matching of the words and sentences given to it.
It is continually supplied bigger and better data – we call it training, but it is no more than lots and lots of data. At the same time, programmers are adding the equivalent of lots more “IF STATEMENTS”

With enough data at hand, it can even generate what some people think is hallucinations – giving wrong answers. You can for example ask it to write a best man’s wedding speech in the style of Shakespeare and it will do it. This is not intelligence, but superb software development.

An LLM is excellent at cross-referencing its knowledge data. It can make links between different keywords very quickly so potentially can come up with possible solutions to a problem that would take an expert longer to analyse.

LLMs are language based, which is why these systems tend to be very poor at maths. They can give word and sentence type answers but traditionally have given wrong calculated answers. It’s an area where software engineers have worked very hard to improve.

Because LLMs are primarily software lead, they tend to be very good at writing software. I use ChatGPT to help write my own PHP or Python code and it has saved me many hundreds of hours. It can even help debug code, find the missing semi-colon or analyse code. This is all due to the amount of very good knowledge in its database and excellent software development. It is getting better and better.