Resources / Interview

Will An AI ‘Writer’ Like GPT-3 Change Your Business ? | OpenAI’s Mira Murati

Join Infinia ML’s Ongoing Conversation about AI

“We saw people using GPT-3 as a satirist, as a customer service agent, as a poet, as a software engineer.”

– OpenAI Senior Vice President of Research, Product & Partnerships Mira Murati

Imagine an almost magical machine that can write anything you tell it to. How would that change your business?

Meet OpenAI, whose long-term goal is “discovering and enacting the path to safe artificial general intelligence.” So, machines that think enough like people to be generally useful but not so much that they want to take over the world or make TikToks.

To be clear, we’re not there yet, but OpenAI’s GPT-3 technology makes it seem like maybe we’re a little bit closer.

The company says that “Given any text prompt like a phrase or a sentence, GPT-3 returns a text completion in natural language.” And boy is it ever returning completed text. In fact, it’s pumping out 4.5 billion words a day.

So can GPT-3 actually write? The reviews say yes…kind of.

The Economist calls GPT-3 “eerily human-like — for better and for worse.” The Verge says it’s “astonishingly powerful but still fighting its flaws.” And MIT Technology Review proclaims it “shockingly good — and completely mindless.”

As a business leader, you may be thinking, “Sure, GPT-3 could be the biggest game changer since the printing press — but what does it mean for me?”

Here are three lessons for leaders from Mira Murati, OpenAI Senior Vice President of Research, Product & Partnerships.

Photo by Wilhelm Gunkel on Unsplash

1. A Machine Can Write Anything

Ok, maybe not literally, but when you’re thinking about business ideas, you can kind of start from that premise.

“You can make it do many different things, so it’s sort of a chameleon,” says Mira.

For instance, the company Viable uses GPT-3 “to extract sentiment and themes in customer reviews [and] chat logs.”

“You might ask Viable, ‘what’s frustrating our customers about the checkout experience,’ and Viable, using GPT-3, will say, ‘Well, customers are frustrated about the checkout flow being too slow.’”

For a completely different use case, Mira points to an application in gaming. Imagine interacting with characters in natural language — and not even the programer knows exactly what they’ll say.

In fact, Mira says, programming dialog is now more about “programming a personality.”

2. No, It Can’t Replace All Your Employees

Especially in high stakes situations where you need to write the right answer, you still need a person to look at what it writes.

“The model is not reliable to be deployed in medical or legal situations directly,” says Mira. “There would always need to be a human in the loop.”

Think of GPT-3 more like a junior analyst or an intern — someone who can take the first pass at a project, but still requires supervision.

“It’s not necessarily replacing the human doing the specific job,” says Mira, “but it’s rather amplifying our intelligence in the same way that mechanical machines have amplified our physical strength.”

3. Anyone Can Use It

“You don’t have to write in the language of the machine” says Mira. “You’re providing your commands in natural language.”

“If you can use such powerful models to do various things without requiring ML knowledge or software engineering expertise, that’s extremely powerful because anyone can use them.”

So what’s next? Mira tells me that machines that combine writing and seeing could be on the horizon.

Can we imagine an AI ‘reporter’ — a drone collects video footage from an event, and then the technology can write a report about what’s happening?

“Eventually,” says Mira. “Perhaps not next year, but I think eventually, we can certainly get there. We’re seeing early signs of these capabilities.”

Thank you for reading! Please subscribe, like, share, comment — give the algorithms what they want.