Many people by now have heard of Chat GPT, the new AI Chat Bot developed by OpenAI. It has set the internet abuzz. There are new features and competitors popping up every day.
Thousands of companies around the world racing to integrate the technology into their existing products, and it’s spawning the creating of hundreds of brand new businesses.
But what exactly is Chat GPT? How does Chat GPT Work? Why are people so excited about Chat GPT?
These are some of the questions that we hope to answer in our brief history of Chat GPT.
What is Chat GPT?
Chat GPT is an Artificial Intelligence Chat Bot developed by OpenAI. It’s taken the internet by storm, and greatly improves on the usefulness of previous chat bots. The advanced technology allows users to log into the Chat Bot and ask it natural language questions, the same as a human would interact with another human.
The differentiator is that Chat GPT provides human levels of accuracy and well written responses. Chat GPT has general knowledge of the world, the internet, and many computer programming languages. The wide breadth of knowledge means that the number of use cases are almost limitless.
The technology behind it has quickly evolved and continues to improve with new version releases.
What is Generative AI?
Chat GPT is part of an emerging field of Artificial Intelligence referred to as generative AI. Generative AI refers to the use of computers to create or generate content. Content could be anything from written text to answer common questions like, why is the sky blue? Or, it could be used to generate images based on a text description like, show me a picture of 3 elves fighting 4 orcs in the middle of Manhattan and receive a photo realistic image of what you asked a computer to generate.
The world of Generative AI is just now taking off, but there are tons of potential efficiency gains for business and the world of digital transformation. Ad copy could write itself, Power Point Slides could suggest tag lines, bullet points, and images. Microsoft Excel or Power BI could answer questions about data like why are sales down year over year?
In each of these example use cases, generative AI could return insights or create content in a matter of moments vs taking a specialist several hours or more to create a similar piece of content.
How do traditional Chat Bots Work?
Chat Bots up until this point in history of largely been unimpressive. They commonly appear in the bottom corner of a website asking if they can help you. If you do reach out to one for help you’re either prompted to enter in your contact information to have a human call you or your get stuck in a frustrating web of questions and non-sensical answers that will quickly remind you of the phone tree you get when dialing into your favorite telephone company.
The reason that most modern day chat bots seem like a phone tree is that they have to be programmed to respond with specific answers to specific prompts. Some advanced ones try to infer which prompt you should get as an answer to your question when the text isn’t a 100% match, but everything largely has to be pre-programmed.
How is Chat GPT different than Other Chat Bots?
Chat GPT differs greatly from other chat bots. Rather than relying on prompts and pre-programmed responses it was developed using advanced Artificial Intelligence and Machine Learning techniques.
What is Artificial Intelligence?
In the simplest terms, Artificial Intelligence is a sub-field of computer science that ultimately aims to give computers the intelligence of a human being. There is an entire field of study that dives into the philosophical questions as to whether this is possible, and what the possible implications are if scientists are successful.
There are a lot of different techniques used in the field of Artificial Intelligence. As computing power has increased and become more accessible with the advent of cloud computing and massive data centers available from all of the major tech companies, the technology has increasingly becoming more accessible.
Artificial Intelligence is the technology behind Self Driving cars from Waymo, Chat Bots like Chat GPT, powers Bing Search, and has countless other applications from developing new medical treatments to bringing the world closer to limitless affordable clean energy.
How does Machine Learning Work?
Machine Learning is a technique used in Artificial Intelligence where a computer is fed a large data set composed of questions and answers.
Questions and Answers are split into two groups when they’re fed into the computer. Questions with answers and Questions with no Answers. The computer then guesses and check its way from a question to a correct answer. This generates an algorithm that it then applies to the dataset that does not have answers pre-determined. A human can then double check the work and give the computer feedback as to whether the algorithm created to bridge question to answer is producing reasonable output.
One of the biggest concerns with Machine Learning is that because a computer is generating the bridge from question to answer, it is sometimes referred to as a black box as people do not necessarily know how the computer is coming up with the answers that it is.
Why is Chat GPT Important?
With the launch of Open AI’s GPT-3, the world began to take notice of this transformative technology. Even though the technology had been around for several years prior, it was the first time that people could really log into Chat GPT and test it out for themselves. By asking questions and receiving well developed, insightful, and useful answers it became easy to see how transformative the technology could be.
The GPT 3.5 and recent GPT 4 models mark significant improvements over each prior iteration. We expect that the pace of improvements will continue to increase, but will be limited by the amount of accurate data that is available to feed into the machine learning model.
We are on the edge of the edge of an AI revolution.
Why does Chat GPT hallucinate with incorrect answers?
Like any new technology, the accuracy of Chat GPT it will improve over time. This is evident in the ability of the leap in capability between Chat GPT 3.5 and Chat GPT 4. There was a massive improvement in the ability of the software to take standardized tests such as the Bar Exam, LSAT, and various medical school entrance exams.
However, part of the challenge with Chat GPT is that it will only be as good as the data set that it is trained on. The internet is full of billions of webpages of information. Much of the internet is full of conflicting information, outdated information, differences of opinion, and some that is completely inaccurate.
Chat GPT is being asked to sort through all of this information and determine which sources are the most credible, and which facts are worth repeating. For example, if you were to search the internet to determine if the world is flat, you would find a wide range of answers, even if the consensus is that the world is certainly round.
Opinions and Past Experience Create Ambiguity Around What is True
During the United States war with Vietnam, the United States Department of Defense commissioned a study by the Rand Corporation to better understand the Vietnamese people and the challenges they were facing in the war.
It was one of the most comprehensive studies ever produced about a topic. It involved thousands of interviews and comprehensive analysis of every aspect of both countries fighting ability from a technological and psychological perspective.
Ultimately, the study failed to come to any specific conclusion or actionable insights because each person who interpreted and summarized the report came up with a different conclusion. The conclusions were grounded in each persons own biases, and past experiences.
In a way, Generative AI like Chat GPT suffer from a similar limitation. Even with a large and comprehensive dataset, answers are subject to interpretation based on the other information that the language model has been trained on. It’s capable of taking into account the context of questions to provide better answers, but will likely not have the insights that a WWII combat veteran or Vietnam Era Politician may have.
Who is OpenAI?
Open AI is the development company behind Chat GPT, along with several other cutting edge Machine Learning solutions that include text generation, image generation, and speech to text conversion. Open AI was founded in December 2015 by a group of prominent tech figures and researchers that include Elon Musk, John Schulman, Greg Brockman, Wojech Zaremba, and Sam Altman.
The mission statement was to ensure that Artificial General Intelligence (AGI) would benefit all of humanity and advancing AI in a manner safe to humanity.
How large was the GPT model?
The first version of GPT, which stands for Generative Pre-Trained Transformer was released in 2018. Transformer technology was a new AI technique for Neural Networks that was first introduced by Ashish Vaswani, [1706.03762] Attention Is All You Need (arxiv.org). The initial model had 117 million parameters and was trained on about 40 gigs of text data.
How large was the GPT-2 model?
GPT-2 was released in 2019, and was trained on a similar sized data set as GPT and had 1.5 billion parameters.
How large were the GPT-3 and GPT-3.5 models?
GPT-3 was the first major increase in training set data and was trained on 570 gigs of text data and 175 billion parameters. GPT3.5 was trained on a similar sized dataset as GPT-3 and similar sized dataset and consisted of a similar number of parameters.
How large is the GPT-4 model?
Chat GPT 4 represents a massive improvement over previous models. The dataset size and number of parameters has not been disclosed. It’s expected that the dataset is significantly larger from prior models with new vision capabilities and the ability to interpret visual input such as images. There are rumors that the model contains close to 1 trillion parameters.
Open AI provides a comparison of the latest version of Chat GPT’s ability to answer questions and answer standardized tests on the OpenAI Research Website.
What is the Future of Chat GPT?
We expect 4 major trends for Chat GPT along with other Generative AI solutions over the next couple of years. Beyond a couple of years out, it’s too difficult to tell how this rapidly developing technology will evolve.
Chat GPT will be seamlessly integrated into most software products
Thousands of Products will Begin Integrating Chat GPT and other forms of Generative AI. It will become ubiquitous and impossible to avoid.
Microsoft and Google have both announced their intentions to integrate Chat GPT style technology into Microsoft Office, the Microsoft Power Platform, and the Google Workplace or G-Suite set of Office Applications. The technology will be so well integrated with major software platforms that most people will begin using Chat GPT without even knowing it.
Chat GPT and other Machine Learning Models will become more efficient
The compute resources required to run large language models (LLM) and run natural language processing (NLP) will go down as machine learning techniques become more efficient. At the same time, specialized hardware will be developed that runs machine learning models. This will make Chat GPT style solutions available offline, and on lower powered hardware resulting in a more useful experience for digital assistants such as Google Assistant and Apple Siri.
The pace of improvement for Chat GPT will continue to accelerate
The pace of technological improvements will increase. This will in part be driven by advancements in software and hardware, but also by the number of people focused on improving the technology. With billions of dollars up for grabs, an AI arms race will permeate all sectors from business, medicine, to national defense, and other scientific endeavors.
As the rate of improvement increases, it will be nearly impossible to keep up with the many ways that generative AI will be able to be used. We imagine that it will be similar to the way that the personal compute, internet, and smart phone revolutionized the workplace. The biggest difference will be how fast AI is adopted as the infrastructure mostly already exists in the form of the previously mentioned technological adoptions by most people.
Advanced AI Solutions and Generative AI Will Become Integrated
We have already seen some instances of image generation and text generation being combined to create advertising campaigns or Power Point Slides, but this trend will continue and become increasingly interesting.
For example, OpenAI,, the creators of Chat GPT have also been working on AI technology called Whisper. Whisper converts people’s speech into transcribed text with astonishing levels of accuracy. It’s able to accurately translate fast talkers, and people with very thick accents. If you’ve ever worked in a call center or drive thru you’ll know how impressive this is.
Now imagine a scenario that a call center is based on Whisper + Chat GPT technology. You could call in using your phone, and the requests could be transcribed in real time to Chat GPT which is trained on company specific policies and procedures to provide you with relevant information.
The only piece missing is the ability for OpenAI to convert text back to speech, but even with the level of technology that exists as show by Apple’s Siri and Google’s Assistant it would be serviceable. Over time it could become transparent and bring into question as to whether you were interacting with a human or an AI.
One question about the scenario above is how much computing power would be required to be able to converse with a well trained chat bot in real time. Improvements in AI efficiency and hardware will likely make this a reality at some point in the not so distant future.
How Can you Prepare for a Future Full of Generative AI?
There will always be holdouts, people that will avoid new technologies on the grounds that it’s dumb, useless, or argue that they don’t want to use it because it could replace their job. Whatever the argument is, or how strong the anti-technology feeling, there is no putting the genie back into the bottle.
In our view, technological changes like these will cause job displacement. There is no way that it cannot. When you can prompt Chat GPT to write a Job Ad and it produces a first draft in 15 seconds, saving someone an hour or two of thought, typing, and editing it’s going to free up time. The time saved could be spent on higher value tasks, which will also become more approachable for a greater number of less skilled workers who are aided by AI.
Being able to execute higher level tasks will still require some training and fundamental knowledge of the topic at hand. This is why we recommend people embrace AI, and begin learning about how to take advantage of it to begin working at a higher level.
If you’re reading this article, you are already ahead of your workplace competition. Begin using tools like Chat GPT and Generative AI in your workplace where possible, and also begin using it to learn new skills.
For example, if you spend a lot of time working with Microsoft Excel, the Microsoft Power Platform opens up a world of new possibilities for reporting and process automation. Some of the tools are approachable but still have a decent learning curve. Chat GPT can help lessen the learning curve, and help people accomplish more than they used to but it will take a lot of hard work and investment of time to improve their marketable skills.