GPT-4 and OpenAI change direction for five companies

SAN FRANCISCO — After testing some of the latest artificial intelligence, companies and nonprofits have agreed on one thing. It means they are already changing the direction of their business.

Five Organizations First to Access GPT-4, the San Francisco Startup’s Latest Product Open AIsaid in an interview that they are reassigning employees, reorganizing internal teams and reassessing strategies in anticipation of technology upending many of their jobs.

Their experience, for better or worse, supports the idea that AI technology could soon fundamentally change some people’s daily lives.

But the organization also says that customizing the technology to its specific needs requires an enormous amount of work. Employees provide daily feedback to the software to train it on terminology and methods specific to fields such as education and finance. Best known for creating the AI ​​chatbot ChatGPT, OpenAI could integrate data from its work into its own models to improve its technology.

In effect, each early tester is a microcosm of what others may experience as access to GPT-4 expands.

Jeff McMillan, Head of Analytics, Data and Innovation, Wealth Management at Morgan Stanley, said:

That’s not true, he said. He said the bank has 300 employees who spend part of their time testing technology using GPT-4.

“We have a team of people literally reviewing all the responses from the day before,” he said.

For Morgan Stanley, the result was a purpose-built chatbot built on GPT-4. It serves as an internal research tool for financial advisor staff. McMillan said the tool has been trained on not only his 60,000 research reports on parts of the global economy, but also his 40,000 other internal documents at the company, and financial advisers to him. have become experts in every financial field they want to find out.

Admittedly, GPT-4 early adopters are not an economic random sample. OpenAI, which became commercial in 2019, has been handpicking organizations over the past few weeks and months.

Critics of OpenAI and its competitors claim the AI ​​sector is benefiting From undoubted hype the past few months. OpenAI was looking for a good example to show when it reached out to non-profit educational organization Khan Academy six months ago, founder Sal Khan said.

“Here’s the context: We’re working on our next-gen model. We want to be able to launch with positive use cases,” he said.

khan academy best known But when OpenAI contacted him, Khan says he has poured resources into creating Khanmigo, a chatbot tutor specially trained on established educational concepts.

“In total, we spent about 100 hours fine-tuning the model to make it behave like a really good tutor,” he said.

“When you look at the cost of tutoring, this can be a very big deal,” Khan added. is.”

When Stripe, a technology company that makes payments software and related products for businesses, got early access to GPT-4 in January, it pulled 100 employees from their regular jobs to hold an internal “hackathon.” and spent each employee an average of one week testing ideas.

Language-learning app Duolingo had access to GPT-4 in the fall, and CEO Luis von Ahn was so enamored with it that he called a meeting at 8am the next morning and immediately changed his job, employees said. I’m here.

“Then he said, ‘Pivot the team,'” said product manager Edwin Bodge. “Since then, we have worked very closely with his GPT-4 and OpenAI teams.”

So far, Duolingo has added a new paid subscription tier for $29.99/month or $167.88/year that gives you access to a conversational chatbot in French or Spanish. We’ve also added an AI bot that walks you through a typical Duolingo lesson, explaining grammar concepts.

According to Bodge, the company created 1,000 to 2,000 word prompts for GPT-4 to power bots. The company does not share prompts upon request.

All the organizations that NBC News spoke with said they were proceeding with some degree of caution given that the AI ​​technology is so new and the potential dangers are unknown. Mike Buckley, CEO of Be My Eyes, a company that makes apps for blind and partially sighted people, tested his app using GPT-4 to bring his version to a wider audience. said he would like to provide I want to be thoughtful and safe.”

“Can we roll this out to the wider community in six to eight weeks? We can, but we’re going to go where the data and use cases lead us,” he said.

The company connects people with low vision with volunteers who can use video calls to explain things around them to app users, such as grocery store product labels, airport directions, and greeting card phrases. The version I did works without a volunteer on the other side, because the AI ​​describes what it “sees” on camera.

According to a video she posted on TikTok, one of the app’s blind spokespeople used it to look up London Underground routes.

“We tried to break it,” Buckley said, adding that his staff had run thousands of tests. I am pleasantly surprised.”

He said his company has never encountered a safety concern with GPT-4, but that it made a mistake. For example, the website conflates a toaster and a slow cooker.