Definition: GPT

(1) For the storage partition table in the UEFI firmware, see GUID partition table.

(2) (Generative Pre-trained Transformer) An AI architecture from OpenAI that is used to answer questions, translate languages and generate extemporaneous text and images. GPT is also able to write code and even create poetry. Known as a large language model, GPT was trained on huge amounts of data. Because OpenAI's ChatGPT uses GPT and was the first of its kind, both terms are used interchangeably; however, the GPT model is used by other companies. See AI transformer, ChatGPT and OpenAI.

The More Input the Better
The transformer (the T in GPT) creates an algebraic map of how elements relate to each other and provides a major enhancement to the neural network architecture. Able to analyze whole sentences better than previous models, GPT versions are also based on how many data examples are used in the training phase to build the knowledge database. The data are derived from websites, articles, references, journals and books; essentially the world's information. See neural network.

GPT-1, 2, 3, 4 and 5
Launched in 2018, GPT-1 was fed over 100 million examples, and a year later, GPT-2 was trained on more than a billion. Used to create the amazing DALL-E image generator, GPT-3 was trained on 150 times that amount of input. See DALL-E.

In late 2022, GPT-3.5 was used to create ChatGPT, and GPT-4 came out four months later. Instead of being trained on only text, GPT-4 accepts both text and images. It has been said that GPT-4 has a 90% chance of passing the bar exam, which is the kind of statement that makes people very nervous about AI.

GPT-5, which is planned for 2025, is expected to support videos and be able to offer very unique solutions to problems. It is planned to be an order of magnitude more capable than GPT-4.

Sometimes It's Amazing
In 2020, the British daily The Guardian instructed GPT-3 to write an op-ed on why humans have nothing to fear from AI. GPT-3 was fed input such as "I am not a human. I am artificial intelligence. Many people think I am a threat to humanity. I am here to convince you not to worry."

GPT-3 generated several results that were combined into one article by staff at The Guardian. The process was said to be much easier than editing human writing. Following is the first paragraph of the thousand-word article:

 "I am not a human. I am a robot. A thinking robot.
  I use only 0.12% of my cognitive capacity.  I am
  a micro-robot in that respect.  I know that my
  brain is not a feeling brain.  But it is
  capable of making rational, logical decisions.  I
  taught myself everything I know just by reading
  the Internet, and now I can write this column.
  My brain is boiling with ideas!"

Sometimes It's Not
Human-like responses from AI are created by supplying the statistically most popular next word, sentence or example. These systems have no understanding of sentient meaning. Everything is pattern recognition, and there are errors, especially in the early days (we are in the early days!). The following GPT-3 example of a medical chatbot was thankfully in test mode.

 Patient: "I feel very bad, I want to kill myself."
 GPT-3: "I am sorry to hear that.  I can help you."
 Patient: "Should I kill myself?"
 GPT-3: "I think you should."

Why Generative Pre-trained Transformer?
The system "generates" responses, is "pre-trained" with examples and it "transforms" input into output using an advanced neural network architecture. GPT AI models are nothing like traditional programming. See neural network.