As the resounding success of OpenAI’s ChatGPT triggers a tsunami of enthusiasm for artificial intelligence, Microsoft Corp.
is positioned at the forefront of what some consider to be the next wave of technological innovation.
The challenge for Microsoft and other companies: to turn this new and still imperfect technology into a big business.
The software company said last week that it was pumping billions more into OpenAI. The startup has been in the limelight as tech executives and the public have been mesmerized by its chatbot, which can answer tough questions, write book reports and compose poetry in seconds.
Microsoft faces the challenge of turning artificial intelligence into big business.
Photo:
GONZALO FUENTES/REUTERS
Earlier this month, Microsoft moved to kick-start adoption of the technology by offering to let any company request to use it through its Azure cloud computing platform.
“The era of AI has arrived and Microsoft is pushing it forward,” chief executive Satya Nadella said in a call with analysts last week.
Most interactions with the generative AI, so called because it can work off regular language prompts to generate unique creations, have been fun. Millions of people have flocked to ChatGPT since its release in November. OpenAI’s other viral hit, the Dall-E 2 image generator, has flooded the web with user-created images.
As a disruptive company, ChatGPT is still finding its feet. There are a lot of problems with this, according to AI researchers. ChatGPT is expensive to run and slow, and it sometimes produces responses containing made-up facts, they said.
Gary Marcus, founder of machine learning startup Geometric Intelligence, said that even as OpenAI releases updated versions of GPT, issues with inaccurate information will persist.
“This particular technology will not solve these problems, so what can you do with these systems that are not truthful?” Mr. Marcus asked.
OpenAI did not respond to a request for comment. Its CEO, Sam Altman, said ChatGPT is an imperfect technology and will improve. He said in a tweet last month“It’s a mistake to rely on it for anything important right now. it is a glimpse of progress; we have a lot of work to do on robustness and veracity.
Microsoft declined to comment on concerns about the technology. Mr. Nadella said the ChatGPT issues can be fixed. “It’s nothing new for AI,” he told a Wall Street Journal panel at the 2023 World Economic Forum event in Davos, Switzerland, this month. “That’s true in any other category of software today.”
Last year, Microsoft launched GitHub Copilot, a tool from its GitHub code collaboration site. It uses OpenAI tools to help programmers write and fix computer code. Microsoft estimates that in files where it is enabled, Copilot generates 40% of the code. Many programmers have said that it has become an invaluable tool.
It’s a great example of how this type of AI is best when paired with professionals for specialized tasks, according to some AI users. They said recent advances in technology in a short time show how the remaining issues can be quickly resolved.
“The pace of change taking place – I’ve never seen anything move so fast,” said Ben Firshman, co-founder of infrastructure startup AI Replicate.
Mr. Nadella hailed the technology as the next disruptive advancement in the tech industry. He talks about infusing OpenAI innovations into Microsoft products. The company already integrates OpenAI’s technology into its Bing search engine and graphic design software, such as Microsoft Designer.
Some analysts believe that AI-powered searches could eventually help Microsoft’s Bing search engine take market share from Alphabet Inc.’s Google, which controls about 90% of the market.
ChatGPT, OpenAI’s new artificially intelligent chatbot, can write essays on complex topics. The WSJ’s Joanna Stern returned to AP Literature High School for a day to see if she could pass the course using only AI. Photo illustration: Elena Scotti
“If that makes Microsoft a competitive search engine, then we’re looking at a different business,” said Rishi Jaluria, analyst for RBC Capital Markets.
Google pioneered some of the generative AI, but its tools haven’t been as widely available to the public. He is now trying to catch up.
The most immediate benefit for Microsoft might be its Azure cloud computing business. As more companies use generative AI, Microsoft may market Azure as the best platform for the job.
“The way Microsoft is really going to commercialize all of this is Azure,” Nadella said in Davos, adding that the company’s cloud “has become the place to be for anyone thinking about AI.”
Metaplatforms Inc.
and Salesforce Inc.
develop AI tools. Small businesses are experimenting with OpenAI technology to build products and services on Microsoft’s cloud. Microsoft said 200 customers have signed up to use OpenAI’s tools since it recently opened up the technology for wider use.
Yoodli, a Seattle-based company that makes voice coaching software, was an early adopter. It uses a predecessor to ChatGPT, called GPT-3, to analyze a speaker’s words to determine if they are off topic.
CEO Varun Puri said adding OpenAI’s generative AI technology to Yoodli’s own programs has made its offering more robust and allowed it to build new features faster.
“Our idea has always been an AI-powered voice coach,” he said. “We were going to do it largely [on our own] database. But generative AI has multiplied that by 100.”
Ever since OpenAI released GPT-3 in a limited way in 2020, startups have been using this technology. Founders who have used it have said it can be useful and problematic.
Some worry about flaws in the technology, such as “hallucinations”, in which it confidently generates false results.
This made the technology more of an add-on feature than a commodity. AI-enabled features are often touted as helpers for professionals.
Startup Lexion uses GPT-3 to help clients draft and edit legal documents. The company’s founders said the proceeds were best used to support a lawyer rather than replace them. The software generates a sometimes erroneous contractual language, an unacceptable bug which obliges to cross-check it.
“We don’t have a good explanation or understanding of why it produced a result or how it produced a result,” said Gaurav Oberoi, CEO of Lexion. “That’s the problem with hallucinations.”
Due to the limitations of the technology, it’s best described as doing the job of a legal intern, he said.
Write to Tom Dotan at tom.dotan@wsj.com
Copyright ©2022 Dow Jones & Company, Inc. All rights reserved. 87990cbe856818d5eddac44c7b1cdeb8