By Adam Pease
OpenAI, the company behind large language AI models like GPT-3, and text-to-image generation tools like DALLE-2, just announced the release of its new chatbot program, based on GPT-3, called ChatGPT.
While ChatGPT demonstrates impressive capabilities as an interactive virtual agent, users have quickly realized its potential to operate as a quasi-search engine as well.
This blog discusses the implications of the news and whether or not models like OpenAI’s have a chance of dethroning legacy search industry leaders like Google.
What Is ChatGPT?
ChatGPT is a chatbot based on OpenAI’s Davinci-03 AI model, its most advanced natural language model to date.
While GPT-3 had some chat-like capabilities, and sometimes demonstrated an impressive ability to quickly resolve user queries or field questions in a variety of different fields, it lacked continuous conversational memory and the ability to convincingly hold a conversation.
For the most part, ChatGPT changes this, offering a much more fluid and naturalistic conversational simulation and supporting a much more impressive memory for conversation history.
It represents a step forward in the increasingly user-friendly suite of OpenAI products and beta tools that delivers on one prediction that Aragon has been making for some time, which is that chatbots will soon achieve a level of natural language understanding and conceptual competence that allows them to excel in fielding many human queries across fields like customer service and more.
Will AI Language Models Replace Search Engines?
While chatbots are a known quantity in the customer service market, what goes less frequently explored is the role they might play in supplementing or even replacing search engines like Google.
When ChatGPT was released, many came to Twitter to announce that it was answering the questions they had Googled better than Google could.
When asked how to approach building a program, for example, ChatGPT was able to provide step-by-step instructions that included specific instructions on which Python programming libraires to use. Google results lacked the same specificity or speed.
Instead of combing a preexisting webpage for answers, users treated the chatbot like a search engine and had it generate entirely new content that met the needs of the query.
This led many to speculate that models like GPT-3 might ultimately be what dethrones Google and shifts us from the era of search engine optimization, towards a new era of search powered by intelligent virtual assistants.
It is an exciting possibility, and one that Google is likely considering itself. However, it must also be stated that while chatbots like this one can deliver very impressive results, they are not always consistent.
ChatGPT also does not currently have the ability to search the web itself, so its knowledge is limited to whatever its most recent batch of training data included, without the ability to dynamically update with the release of new information in real-time.
Bottom Line
Despite these uncertainties, the release of ChatGPT certainly seems to herald an interesting future for search engines.
It remains to be seen how such a tool could be productized, especially considering the heavy reliance status quo search engines have on ad revenue, but large language models present exciting possibilities for revolutionizing the way we ask and answer questions digitally.
SEE ADAM AT TRANSFORM 2022!
Aragon Research’s analysts discussed, ‘The Top Ten Technologies for 2023 and 2028″ on Thursday, December 8th at 10 AM PT / 1 PM ET.
To see the full list on-demand, register here!
