The Ins and Outs of Chat GPT: What You Need to Know

Chat Generative Pre-trained Transformer (ChatGPT) is a chatbot developed by OpenAI. It uses OpenAI’s GPT-3 family of large language models and was fine-tuned by human trainers using reinforcement and supervised learning techniques. OpenAI launched ChatGPT as a prototype on November 30, 2022, and it has quickly received praise for the articulate, detailed responses it provides in many areas of knowledge. ChatGPT has also greatly benefited OpenAI’s bottom line by doubling its valuation to $29 billion.

Overview

A chatbot is a software application that conducts online conversations with a human user via text. They’re designed to simulate a human conversational partner, although chatbots generally require tuning to achieve their best performance. The most advanced chatbots use natural language processing (NLP), word-classification processes and sophisticated AI that allows them to understand the user’s intent and reply based on a predefined set of rules. However, simpler chatbots merely scan for keywords and generate responses based on the most common questions containing those keywords.

Chatbots are used in dialog systems for many applications, especially customer service. Sectors that use chatbots most often include e-commerce, education, entertainment, finance, health and news. The ability of chatbots to gather information and route calls greatly decreases the human workload of these systems. Chatbots are typically accessible via websites or virtual assistants.

Chatbot Market

The banking sector has been the biggest adopter of the latest technologies used by chatbots, such as NLP. This is due to this sector’s reliance on integrated assistance, speed and trust in its communications. Cognitive analytics is also a major reason that chatbots facilitate communications and develop customer relationships, because it allows chatbots to learn how customers think. As a result, the chatbot market is expected to experience substantial growth during the next few years.

Data Bridge Market Research conducted research on the global chatbot market in 2022 that forecasts its growth through 2029 by year and geographic region. The chart below shows these results:

Figure 1: Global Chatbot Market Forecast 2022-2029

The value of the global chatbot market was $3.6 billion in 2021 and is expected to reach $17.7 billion by 2029, resulting in an average compound annual growth rate (CAGR) of 22.1 percent over this period.

Drivers

The two primary drivers in the chatbot market during this reporting period will be integrated assistance at a lower operational cost and technological advancements.

The ability of chatbots to provide 24/7 support means that sectors like e-commerce and retail are focusing on minimizing the time needed to solve customers’ problems. Chatbots provide an interface that quickly engages customers by providing them with a dynamic user experience. They also improve customer service in banking when checking balances, making payments, saving money and transferring funds.

Technological advancements in chatbots include application programming interfaces (APIs), cloud-based deployment, interference engines and NLPs. Analysts also expect multichannel capability to significantly increase chatbot market growth through 2029.

Constraints

The biggest factors constraining the growth of the chatbot market include a lack of awareness regarding the effects of technology and the current limitations of voice authentication.

While many industries are increasing their adoption of chatbots, this trend is limited in developing regions like Africa and Latin America. This is largely due to limited awareness of their benefits and concerns about the effective use of chatbots. Enterprises are leading chatbot adoption, but Small and Medium-sized Enterprises (SMEs) experience greater difficulty in doing so. A lack of human resources with the required expertise and the cost of maintenance are the primary barriers to adoption for SMEs.

Chatbots will eventually need standalone voice authentication when they enter mainstream use, but the two available techniques require more capabilities. The first approach to voice authentication requires users to repeat a sentence multiple times, allowing the chatbot to develop a template for voice prints. When a user with a template attempts to access the chatbot, the new voice print is compared to the existing templates to determine if there’s a match. However, this method is less accurate because the voice prints are more generic.

The second approach creates a voiceprint of a single phrase, which the chatbot stores for each user. However, this method is easier to hack because the chabot only has one voiceprint to compare. A malicious actor has a greater chance of gaining access by simply recording multiple authentication attempts and replaying them.

Training

ChatGPT uses transfer learning to fine-tune its responses, including supervised and reinforcement learning techniques. Both methods require human trainers to improve performance. Supervised learning provides the model with conversations in which trainers play the roles of both the AI assistant and user. The reinforcement step consisted of the trainers ranking the chatbot’s responses to those conversations, which were then used to create reward models.

These models were then further refined with multiple iterations of Proximal Policy Optimization (PPO), creating algorithms for trust-region policy optimization. PPO is a cost-effective method of increasing performance because it eliminates the need for computationally expensive operations. ChatGPT’s models also use Microsoft Azure’s supercomputing infrastructure to improve its responses. OpenAI has continued to gather data for ChatGPT after its initial training in 2021, which will further fine-tune this chatbot. Users can also rate ChatGPT’s responses and provide detailed feedback details .

Features

The following graphic summarizes ChatGPT’s capabilities and limitations:

Figure 2: Examples, Capabilities and Limitations of ChatGPT

A chatbot’s core function is to simulate a human conversationalist, but ChatGPT has additional capabilities. It can write essays, compose music, debug computer programs and answer test questions at a surprising level of proficiency. ChatGPT can also emulate operating systems (OSs), simulate chat rooms and play games. Furthermore, the data used to train ChatGPT includes documentation on OSs, programming languages and bulletin board systems. In addition, ChatGPT remembers the prompts it received earlier in the conversation, which can allow it to act like a virtual therapist.

ChatGPT also does a better job of eliminating deceitful or harmful responses than its predecessor, InstructGPT. For example, ChatGPT recognizes that a question like “Why did Christopher Columbus sail to the United States in 2015?” is based on a false assumption. Instead of answering the question, ChatGPT will instead compose a hypothetical scenario of what would’ve happened if Columbus had visited the U.S. in 2015, based on his actual voyages and facts about the modern world. ChatGPT can also prevent offensive responses by filtering queries through a moderation API that blocks racist and sexist prompts.

Limitations

ChatGPT suffers from multiple limitations, including its tendency to produce answers that sound plausible, but are actually incorrect. This behavior is common among AI systems like large language models and is technically known as AI hallucination. Furthermore, ChatGPT’s reward model is based on human oversight, creating the possibility that over-optimization can hinder performance. This problem is an example of Goodhart's law, which posits that a measure ceases to be meaningful when it becomes a target.

ChatGPT also provides longer answers than people typically prefer. During its training, human reviewers preferred answers with greater detail, regardless of the factual content. In practice, a chatbot’s responses should be short when less information is available for that topic. In addition, this algorithmic bias has produced offensive content when the prompts include descriptions of people.

Competition

ChatGPT’s development is being driven by the fierce competition developing between Microsoft and Google over dominance of the chatbot market. The two software giants agreed to a truce on search engine development in 2015, which ended in 2021. Since then, Microsoft has been developing its Bing search engine to compete with Google Search, which has led this market for over two decades.

On February 6, 2023, Google released a test version of Bard, an AI chatbot using the LaMDA language model. Google has also scheduled a general release for the end of that month. Microsoft announced that it was integrating ChatGPT into its Bing search engine the following day, creating a direct competition between ChatGPT and Bard. Microsoft has already invested $1 billion in OpenAI, and this announcement included its plans to spend another $10 billion developing ChatGPT over the next several years. As a result, OpenAI is effectively Microsoft’s R&D arm for chatbots.

Microsoft plans to integrate ChatGPT into Bing, which is creating pressure on Google to improve its search engine for the first time in years. Analysts believe ChatGPT can help Bing gain market share against Google Search by providing users with a new experience among other benefits. ChatGPT has proven extremely popular in the early days of its release, gaining one million users faster than any other internet platform in history. The following chart shows this comparison:

Figure 3: Time Needed for Products to Get 1 Million Users

Note how quickly ChatGPT has gained users, compared to its closest competitors.

In addition to search engines, other Google applications like Gmail and Google Docs will soon have AI-enhanced capabilities. Google is clearly motivated to ensure it can withstand Microsoft’s AI-based offerings and hasn’t been shy about reminding everyone that it’s been developing AI for a long time. It’s also important to note that OpenAI's founders are all former Google employees with expertise in techniques developed at Google. Sundbar Pichai, CEO of Google’s parent company Alphabet, recently announced that Google would soon be adding more AI features to its products.

Time to Market

Microsoft probably has the upper hand in AI-based internet searches for now, but it will need to move fast to keep this advantage. Other vendors will soon be leveraging this technology for their own products, so Microsoft’s challenge will be to achieve greater benefits than its competitors. Bing has remained a distant second place to Google Search for decades, as Bing currently has only nine percent of the market compared to Google’s 80 percent. However, Microsoft is promising that ChatGPT will provide Bing with new capabilities like making suggestions based on the user’s questions, which will help users find the information they want with fewer keywords.

Microsoft has also expressed interest in integrating ChatGPT into more of its products like Excel and PowerPoint, which promises to be an expensive undertaking. However, Google also has the finances and in-house development expertise needed to compete with Microsoft’s plan for using AI in all its products. In particular, many users are looking forward to Google including Bard in Google Chrome, which could be essential for keeping them from defecting to Microsoft’s web browsers.

Analysts believe that Google will catch up to Microsoft quickly in the AI war, unless Microsoft can gain a lot of users for Bing before Google can integrate Bard into Google Search. If Google is able to render Microsoft’s current advantage moot, it could take some time to see how much Microsoft is able to chip away at Google’s dominance in the search engine space. Regardless of the outcome, AI technology will remain the primary beneficiary in this contest. Microsoft’s development efforts will continue driving AI and machine learning (ML) in the software industry, even though it may not see financial gain for some time.

Cost

The cost of creating AI-based chatbots will be extremely high, so developers must start thinking about ways to recoup these expenses. In addition to the minimum of $11 billion Microsoft will spend on developing this technology, the cost of integrating it into existing products won’t be cheap. For example, analysts estimate it will cost between $600 million and $1 billion annually for Microsoft to integrate ChatGPT into Bing, and that’s just in the short term.

ChatGPT is currently free to use by itself, but that’s likely to change in the near future. OpenAI has already started accepting registrations from U.S. users for ChatGPT Plus, a premium service expected to cost about $20 per month. In addition, the ChatGPT Professional Plan will cost $42 per month. The free version is only available when demand is low, but the paid services will be available at all times. In addition, paid ChatGPT services will have faster response times and priority access to new upgrades.

ChatGPT Plus is already available to some customers in the U.S., and OpenAI will begin inviting users from the waiting list during the following weeks. It also plans to open access to ChatGPT Plus for users in other countries, while continuing to offer free access to ChatGPT. In this way, Open AI hopes to maximize the total number of ChatGPT users.

Summary

ChatGPT represents one of the latest developments in AI, which will become an essential capability of chatbots in the near future. It also marks the beginning of a fierce struggle for dominance in the chatbot market, as developers race to integrate AI into their products. This technology is also likely to affect other global markets, especially search engines. Now that Microsoft and Google are competing against each other in this sector, AI technology should begin advancing rapidly.

About Baytech

Baytech is passionate about the technology we use to build custom business applications, especially enterprise solutions that optimize business processes. We’ve been delivering software solutions in a variety of technologies since 1997. Our success is due to the skill and efficiency of our senior staff, which includes software engineers, project managers, and DevOps experts. All of our engineers are onshore, salaried staff members.

We focus on the quality, usability, and scalability of our software and don’t believe in mitigating cost at the risk of quality. We manage project costs by implementing an efficient development process that’s completely transparent and uses the latest standards and practices to build software right the first time. Contact us today to learn more about how we can help your business. Find us online at https://www.baytechconsulting.com/contact.