This website evaluated various chat bots using artificial intelligence. There are many competing AI.models and the most famous is Generative Pre-trained Transformer, often shortened to GPT. GPT is developed by a company named OpenAI and Microsoft is heavily involved in this company.
The purpose of improving and evaluating chat bots based on various AI-models is to compile vast amounts of information and upon request present relevant parts of this to physicians. Most of the external sources we provide to our chat bots are aimed for physicians. Hence, their replies may often be too complex for lay persons. They works best with a single question where a single reply is delivered. They usually work less well if you try to have long discussions, especially not in religion, politics, morale or ethics. Our chat bots are named Petra 1, Petra 2 and so on.
The first version of GPT was launched in 2018. Version 2 was launched in 2019 and version 3 in 2020. The first model being publicly available was version 3.5 launched in 2022. Version 4 was released in early 2023. The different models are compared below*:
| Model | Parameter count** | Training data | Release date |
|---|---|---|---|
| GPT-1 | 117 million | 4.5 GB of text, from 7000 unpublished books of various genres. | June 11, 2018 |
| GPT-2 | 1.5 billion | 40 GB of text, 8 million documents, from 45 million webpages upvoted on Reddit. | February 14, 2019 (initial/limited version) and November 5, 2019 (full version) |
| GPT-3 | 175 billion | 499 Billion tokens consisting of CommonCrawl (570 GB), WebText, English Wikipedia, and two books corpora (Books1 and Books2) | May 28, 2020 |
| GPT 3.5 (Chat GPT) | 175 billion | Undisclosed but likely to be similar to version 3. | March 15, 2022 |
| GPT 4 | Undisclosed but estimated to be 1760 billion | Undisclosed | March 14, 2023 |
*Information from Wikipedia.
**Parameters in GPT are the learned variables that encode the knowledge acquired during the training process to make relevant predictions. Example of parameters used in GPT:
- Weights: They are the most fundamental parameters in GPT that model learns from the training data. For instance, if GPT sees the word “cat” followed by the word “meowed,” it will assign a higher weight to this word pair. Next time, the model will more likely predict the word “meowed” after the “cat.”
- Bias: This parameter is used as an adjusting factor to tune the prediction of an entire layer to the more accurate side. Because of this reason, it is the same for a specific layer and applied layerwise.
- Learning rate: This parameter suggests how much the biases and the weights have to be adjusted in order to reduce the error and make the results more accurate. This process of reducing the error is done during backpropagation.