SAN FRANCISCO — After testing the latest in artificial Intelligence, businesses and non-profit groups have come to an agreement: It has already changed the course of their operations.
Five companies that had access to GPT-4, the latest product by San Francisco startup OpenAI said that they were reorienting their internal teams, reevaluating their strategies and reassigning employees in preparation for the technology changing a lot of their work.
These people have confirmed that AI technology could very soon change their lives for the better.
However, the companies also stated that the technology was difficult to tailor to their needs. Employees provide daily feedback to the software in order to train it with terminology and methods relevant to their respective fields. OpenAI, which is best known for its AI chatbot ChatGPT can then incorporate the data from this work into its own model to improve its technology.
Each of these early testers is, in effect, a microcosm for what other people might experience as GPT-4 access expands.
“There is a perception that you can plug into these machines, and they will give you all the answers,” Jeff McMillan, head analytics, data, and innovation at Morgan Stanley’s wealth management division, said.
He said that this is false. He stated that 300 bank employees spend some time testing their technology using GPT-4.
He said, “We have a team that literally reviews every response from the previous day.”
The result was a chatbot that Morgan Stanley created with GPT-4. It serves as an internal tool for financial advisers and is now used by the firm. McMillan stated that the tool can be trained on not only 60,000 research reports about parts of the global economic system, but also on 40,000 internal documents from the company. This makes it an expert in any financial topic that a financial advisor might need to know.
GPT-4’s early adopters aren’t random, it is certain. OpenAI, which was made for-profit in 2019, carefully selected the organizations over the last weeks and months.
OpenAI critics and competitors claim that the AI industry has profited from the unskeptical hype in the past few months. OpenAI was seeking positive examples to demonstrate its intentions when it reached out to Khan Academy six months ago, the founder of the non-profit educational organization Khan Academy.
He stated that the context was “We’re going to work on a next-generation model; we want it to be able launch with positive use cases.”
Khan Academy is most well-known for its YouTube videos. However, OpenAI reached Khan to inform him that Khan has invested resources in Khanmigo, a chatbot tutor who is trained in teaching concepts.
He said that the group spent around 100 hours tuning the model to make it behave like a good tutor.
Khan said that tutoring can be expensive. It’s like having a tutor, tutor, or graduate student that you can talk to immediately.
Stripe, a tech company that creates payments software and related products, stated that it took 100 employees out of their regular jobs in January to get early access to GPT-4 and assigned them to an internal hackathon in which they spent one week testing out new ideas.
Duolingo is an app that helps people learn languages. In the fall, employees were able to access GPT-4. CEO Luis von Ahn was so impressed with the app that he called a meeting at 8 a.m. on the next morning and changed people’s jobs immediately.
Edwin Bodge, a product manger, stated that “He, after this, said, ‘Pivot Your Team.'” “We’ve been working closely with GPT-4 since then and the OpenAI team.”
Duolingo now offers a paid subscription that costs $29.99 per month, or $167.88 annually. This allows you to access a chatbot in French and Spanish. Duolingo has also included an AI bot that will help you understand grammatical concepts as you go through the lessons.
Bodge claims that the company has created prompts of 1,000 to 2,000 words for GPT-4, which power the bots. The prompts were not shared by the company upon request.
All the organizations that spoke to NBC News stated they are taking precautions because AI technology is still new and there is no way to know if it will prove dangerous. Mike Buckley, CEO at Be My Eyes (a company that creates an app for people with low vision or blindness), said that he would like to see a more tested version of the app with GPT-4 in more hands but that “we want to be thoughtfully safe.”
“Could this be launched more widely to the community in six to 8 weeks?” He said it was possible but that we would go wherever the data and use cases lead us.
The company connects low-vision people to volunteers who can, via video, describe what they see around them. This could include a product label at a grocery store, directions through an airport, or even the greeting card text. GPT-4 version works without the need for a volunteer because the AI “sees” what it sees with its camera.
According to TikTok, one of the app’s blind representatives used it to find directions to the London Underground subway system.
Buckley stated that they had tried to break it, and his staff conducted thousands of tests. “We have been amazed at the results of our efforts to hammer the technology for several weeks.
Although he said that his company had not encountered any safety issues with GPT-4, he acknowledged that it made mistakes, such as mixing up a toaster and a slow-cooker for a website.