• by Laurie Sullivan , Staff Writer @lauriesullivan, December 13, 2022

Microsoft, which has dabbled in its own chatbot creations throughout the years, invested $1 billion in OpenAI in 2019, and reportedly conducted advanced talks for new funding as resent as October 2022.

OpenAI is known for its chatbot model called ChatGPT, which is trained to have conversations and respond to complex queries with intelligence and accuracy.

“The dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests,” according to the company’s blog posted November 30, 2022. “ChatGPT is a sibling model to InstructGPT, which is trained to follow an instruction in a prompt and provide a detailed response.”

ChatGPT crossed more than 1 million users in five days, says Aleksandr Volodarsky, founder and CEO of Lemon.io, a Ukrainian remote staffing company for engineers. In the tweet, he compared the growth to that of Netflix, which took 41 months to surpass 1 million users, while Facebook took 10 months to reach that level and Instagram took 2.5 months, he wrote.

Rakshit Lodha, product manager at INDmoney, used ChatGPT to build a Twitter bot without any programming knowledge.

“I started using it instead of Google for a lot of research,” he wrote in a post. “Why? Simple, because it just gave me an answer instantly. Slowly but surely, I was hooked on it. And I started thinking of how I could use it to build something cool that I could later share. That’s when I decided to build a Twitter bot. Why a Twitter bot, though? I don’t know, it sounded cool.”

OpenAI trained the model using Reinforcement Learning from Human Feedback (RLHF).

The model initially was trained using supervised human AI trainers who provided conversations in which they played both sides — the user and an AI assistant. Trainers were given access to suggestions to help them compose responses.

I began to think about how Microsoft might use ChatGPT or a similar technology to further search and advertising across its platforms. While this is complete conjecture on my part, the chatbot can do advanced things that are offered through voice search.

It can help people write code even when the code is a foreign language. It can help people learn, similar to the way search engines like Bing or Google do, through queries and answers.

Will ChatGPT become the new search engine of the century?

There are challenges based on bias in search. OpenAI acknowledges the challenges and limitations facing ChatGPT because much of the information provided by the chatbot comes from search engines.

OpenAI acknowledges that ChatGPT sometimes writes plausible-sounding but incorrect or even nonsensical answers. Fixing this issue is challenging for several reasons. Here are some reasons that the company states, but marketers can read about others here.

• During training there is currently no source of truth
• Training the model to be more cautious causes it to decline questions that it can answer correctly
• Supervised training misleads the model because the ideal answer depends on what the model knows, rather than what the human demonstrator knows.

Bloomberg points out that it can help students cheat on their essays, as it lacks ethics. It observes how people often cite authoritative voices when building an argument, so it can fabricate believable quotes to augment its arguments. And it can write a company earnings report in the style of a business publication. And while it’s very believable, none of those facts and figures are true.