6 March 2023
Chat Generative Pre-trained Transformer, known commonly as ChatGPT is a chatbot introduced in November 2022 by OpenAI. The free-of-charge prototype version gained a lot of attention by the end of last year and, as a result, was valued at US$29 billion. It’s supposed to process and generate a human-like text without having an access to the Internet. That being said, it cannot browse the net for anything and it can only provide information it has been trained on, which didn’t stop it from obtaining its first million users within the first 5 days after launching.
ChatGPT is what it is thanks to supervised learning combined with reinforcement learning, both being rooted in the machine learning field. The supervised learning aspect provided ChatGPT’s algorithm with the ability to analyze given data and create new examples based on the correlation between a feature and a label. The quality of the algorithm (and its performance, obviously) is measured through generalization error – the measure of how the algorithm actually is when operating on previously unseen data. This is why sometimes users got an error message, which will improve over time, also because of the reinforcement learning part.
Reinforcement learning is heavily based on interactions with the environment. In the ChatGPT case – the users are the environment and the hype around the product made the learning conditions more than perfect. ChatGPT is learning by collecting data, providing some outcome, and checking if the person who asked for it is happy with the result. The same type of learning is used for teaching AI how to play chess and other games.
Absolutely not. Comparing ChatGPT (in its current form) to Google search doesn’t make a lot of sense, since the point of Google search engine is to search the Internet which ChatGPT cannot do. If it wasn’t provided specific information on a subject it cannot generate any text about it. The point of using Google search is to find sources and potentially use them to create an original piece. ChatGPT can check if a text given is correct grammatically, but that’s it.
There was a short burst of interest around the question if it can write a thesis or at least a piece of essay that can be used for finals and it kind of can, except with no footnotes or references, which makes it useless. The risk of plagiarism is another issue since ChatGPT is blending things together rather than creating anything groundbreaking.
In January 2023 however, Microsoft announced “a new multiyear, multibillion-dollar investment with OpenAI” who is the maker of ChatGPT and in February a paid pilot subscription has been launched, so the future sure looks exciting.
ChatGPT has definitely spurred one thing – the global AI race. In February 2023 Google has introduced its product, and later the same month – Chinese companies informed about their own versions which is not surprising since China is generally known to restrict some of the Western pages for their citizens and provide some sort of counterparts. One of those versions, however, crashed after a few minutes and it’s not accessible to the general public at this time.
There are plenty of tasks that can be done by ChatGPT which will give more time for your employees to perform tasks that only humans can succeed at doing. ChatGPT can deal with things like appointing meetings or data entry. It can also generate some of the basic messages that you usually send to your business partners or clients. For the same reasons, however, it might be capable of participating in scams or phishing attacks, by asking victims for sensitive data or impersonating their family members.
The point is to enhance the customer experience and using AI is one way to go. The future will show how far it’s going to take us.
Author: Andrzej Wodnicki, Managing Director at ITSG Global