Skip to content

Let’s ChatGPT: The sky-high expectations of AI chatbots

Over one million people talked to artificial intelligence (AI) system ChatGPT in the four days since it was released for public testing last week. Research company OpenAI created ChatGPT to have conversations with people and “answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.” It was a smashing success.

Tech enthusiasts like Shital Shah collected their favorite cases of ChatGPT giving articulate answers to some shocking prompts, like suggesting the agenda for an all-hands meeting, writing software code in a specific programming language, or explaining quantum theory to a child in the style of rapper Snoop Dogg. To borrow from that last answer, ChatGPT is a wild ride but it’s a cool one too.

Of course, some people found ChatGPT lacking. It seems the system can sometimes give false answers with utter confidence. These limitations are listed right on top of the box where you enter your ChatGPT prompts, where the user is warned that the AI chatbot “May occasionally generate incorrect information,” and “may occasionally produce harmful instructions or biased content.”

After those warnings, it shouldn’t be a surprise that the system may fail, but the technology is so promising that expectations are sky-high. OpenAI CEO Sam Altman summed it up by saying that “ChatGPT is incredibly limited, but good enough at some things to create a misleading impression of greatness,” and he added, “fun creative inspiration: great! reliance for factual queries: not such a good idea.”

At the moment, a more pressing concern for AI technology in general is how to pay the humans producing the information on which new systems like ChatGPT train their skills. Someone wrote the original meeting agendas which ChatGPT used to recommend what to cover on yours: if they don’t get paid, that might be the last agenda they ever put out. This is a problem for systems producing automated art too, like OpenAI’s (also excellent) DALL-E 2.

Perhaps that’s only a problem if you see ChatGPT as the product instead of a technology helping produce a more specific benefit that clients are willing to pay for, like publishing software code. Software publishing service GitHub has an AI feature of the same generation as ChatGPT called Copilot. Former GitHub CEO Nat Friedman has said that he expected a lot more companies to introduce AI features like Copilot but has since realized that entrepreneurs are only now starting to imagine how to apply these new AI capabilities.

If that were the case, we should start seeing powerful AI features in a lot more products of every kind, and even some new products nobody has ever thought of yet. People have talked ChatGPT into doing a lot in a very short period. Given more time, we could see some exciting topics come out of this chat. It’s only a matter of thinking up a new prompt. Time to try your own.

Your own personal Google

We have all been entering our prompts into a different system for some time now: Google. Startup founder David Petersen tried entering the exact same prompt at Google and ChatGPT, “please list 15 important inventions,” and got the usual list of blue links that you’d expect on one side, and a proper enumeration with all the requested details on the other. He reflected ChatGPT might prompt some changes at Google to remain competitive.

Google has long invested in AI and parent company Alphabet’s research lab Deepmind has operated since before OpenAI. But there might be someone else to spur new competition: Microsoft has invested in OpenAI and Altman, OpenAI’s CEO, has said that Microsoft Azure is behind much of the company’s progress. One practical move you could see soon: SEO (Search Engine Optimization) expert Lily Ray says Google will likely tweak its algorithm again soon to deprioritize automated text in favor of sources showing real expertise, authority, and trust, or E-A-T, as they’ve been doing for some time.

Some of the latest AI models run on Microsoft Azure

Productive Chat

Does this mean that you will again have to deal with supposedly smart systems to get answers from anyone? Not necessarily believes Alan Masarek, CEO of Avaya, the leading call center software provider (and a Verb customer). Rather than tolerating disruption, he said, “many thousands of our customers want to augment voice-only contact center functionality with new digital channels like chat, text and social.” Instead of replacing customer service agents, companies are asking Avaya for help to make it easier for agents to team up with other people in the organization to solve customer problems, he says. It that scenario, an AI system that makes everyone involved more productive could be a welcome chat.