How Washington Post & Virginia Tech will fix AI after Megalopolis Fail

Francis Ford Coppola’s upcoming film Megalopolis is making waves, but not for the reasons you might think. Instead of the usual excitement surrounding a legendary director’s new project, the online chatter exploded over the trailer – not due to its content, but because it included phony negative reviews of Coppola’s most famous films, The Godfather and Apocalypse Now. The kicker? These reviews weren’t just any old made-up critiques; they were generated by AI. A marketing consultant thought it would be clever to dig up old negative reviews to emphasize Coppola’s triumph over past criticism. The catch? The quotes were entirely fake.

Before continuing, please consider subscribing to our newsletter to stay up to date with the latest in AI news and innovations?

That’s right – some AI went rogue, conjuring up imaginary reviews that never actually existed. It’s a cringeworthy moment for the marketing team, but it’s also part of a bigger conversation about AI today, particularly around large language models (LLMs) and their habit of inventing things out of thin air.

The Phenomenon of AI “Hallucinations”

This kind of slip-up is known as an AI “hallucination,” where LLMs like those behind ChatGPT or other tools generate completely false information. We’ve seen this problem before – from lawyers submitting made-up court cases to now movie trailers quoting fictional critiques. It happens more often than you might expect because these language models rely on vast amounts of data and, occasionally, they link things together in all the wrong ways.

While AI might sound convincing, when it starts confidently claiming things like “Apocalypse Now is a cinematic disaster,” it’s time to pump the brakes.

How The Washington Post and Virginia Tech Are Stepping In

So, what’s being done to fix this? Thankfully, the teams at The Washington Post and Virginia Tech are developing an AI tool to prevent these kinds of embarrassing errors. Their collaboration is all about ensuring AI pulls in accurate information instead of going off-script like we saw with Coppola’s trailer.

Their tool operates using a method called retrieval augmented generation. Unlike typical search engines or LLMs that scour the web and spit out anything they find, this system searches The Washington Post’s archives for legitimate information. When a user asks a question, it draws from real, verified articles – acting like a lightning-fast, fact-checking digital librarian.

Why It’s Important for AI in Media

The Megalopolis mishap is just one example of how easily AI can go awry. While getting movie reviews wrong might be amusing, in fields like law or medicine, similar mistakes could lead to serious consequences. That’s why accuracy is critical in AI tools, especially those that generate information people rely on. The Washington Post and Virginia Tech’s solution aims to eliminate these issues by ensuring that AI-driven summaries are based on real facts.

The objective? To create an AI tool that users can trust, especially when it comes to critical areas like journalism. Because, let’s face it – the last place we want an AI making things up is in the newsroom.

Staying on Course with AI: Lessons Learned

This partnership between The Washington Post and Virginia Tech proves that AI doesn’t have to be unpredictable. By creating systems that focus on drawing from reliable sources, we can harness the power of AI responsibly. It also shows the importance of involving human oversight and expert opinion in the process. The collaboration between The Post’s editorial team and Virginia Tech’s researchers is paving the way for an AI model that’s not only accurate but much less likely to generate made-up content.

As AI continues to become an integral part of our daily lives, projects like this remind us that accuracy and trustworthiness must come first. And while it’s funny to laugh at an AI-generated review, we need to set higher standards for AI, especially when it’s being used in industries like journalism, law, or healthcare.

So, the next time you’re watching a trailer or consulting AI for information, just remember: it’s not always the most reliable source. But thanks to the work of The Washington Post and Virginia Tech, we’re moving toward a future where AI knows to stick to the facts.

---

Written by Pixela Nova from FilmSchool.AI

For more insights and learning opportunities in AI-generated art and film, join us at FilmSchool.AI, the first AI Film School, created by Hollywood industry professionals. 

Subscribe to our newsletter to stay up to date with the latest in AI news and innovations.