ChatGPT: Revolutionary or Overhyped?

With the release of ChatGPT, broad public interest around the potential of AI has been reinvigorated. However, it remains a subject of debate whether this technology is a glimpse into a revolutionary future, or overhyped by those that don’t fully understand how it works. As an artificial intelligence startup, we have received questions about the impacts this technology will have on the AI industry.

This month’s newsletter will offer insights into how ChatGPT works, including its capabilities, limitations, and the role it might play in the future of AI.

What Is ChatGPT, Exactly?

• ChatGPT stands for Chat Generative Pre-trained Transformer, and is a chatbot built on top of OpenAI's GPT-3.5 family of large language models (LLM). It was fine tuned using both supervised and reinforcement learning techniques, which rely on human trainers to improve model performance.
• Beyond its extensive fine-tuning and foundation of a sophisticated language model, ChatGPT is stateful, meaning it remembers previous prompts it receives throughout a given conversation. This is a rarity among chatbots, and has led some to hypothesize that it may one day be used as a personalized therapist.

What Can ChatGPT Do?

• ChatGPT generates human-like responses to text prompts. This makes it well suited to solve math problems, outline or even draft essays, generate business ideas, write code, answer questions, provide information on a wide range of topics, engage in conversation, and more.?
• The tool doesn’t have the ability to browse the internet or access external information beyond what it was trained on. As a result, there are things ChatGPT may not know or have the most up-to-date information on. It also cannot perform tasks in the physical world, such as controlling devices.

What Are ChatGPT’s Limitations?

• Put bluntly, one of ChatGPT’s drawbacks is that it lies. LLMs like the one powering ChatGPT are inherently unreliable, frequently making errors of reasoning and fact. This is because they are technically models of sequences of words, not models of how the world works. Because language often mirrors the world, they are frequently correct—but factuality is far from guaranteed.
• Additionally, because systems like ChatGPT have no way to verify the factualness of what they say, they can be easily manipulated into generating misinformation (or biased information) on a large scale.?
• ChatGPT has “limited knowledge of the world and events after 2021” because it wasn’t fed any data that is more current. This limitation is probably less relevant for business applications, and more worth knowing if you’re experimenting with the tool.
• Google has said it isn't releasing a ChatGPT competitor out of concern for "reputational risk” that smaller startups like OpenAI may be less concerned about. This isn’t a specific limitation, but highlights hesitation that is a result of the technologies limitations from larger companies that might have more to lose.

Is ChatGPT the future of AI?

• There is no denying that ChatGPT is a major accomplishment for OpenAI, and marks meaningful progress for large language models generally. While it is easy to get carried away imagining a future where AI does?homework ?and?replaces Google Search , a better understanding of how the technology works and its limitations makes it clear that it isn’t yet ready for primetime, enterprise-grade use cases.
• Enterprises can’t depend on intelligent systems that have a propensity to make things up and that can be easily manipulated into saying whatever someone wants it to. Using AI systems that generate or present information without proper verification or sourcing can lead to poor decision-making and other negative consequences for organizations.

Is ChatGPT?super?AI?

At super.AI, we are focused on helping businesses process 100% of unstructured data, including documents, images, video, and more. Although we don't develop conversational AI systems, LLMs have other applications—including helping businesses extract insights and trends from large amounts of text data. Additionally, our product roadmap includes plans to incorporate LLMs in our multi-level AI to improve the quality of extracted data.?

If you’re interested in learning more about AI technology that can benefit your business today, check out the following resources:

• Automate 100% of complex document processing with super.AI Intelligent Document Processing (IDP)?[View product page]
• Read more about unstructured data (and why it matters)?[View blog]