A Reddit user who claimed to be a whistleblower from a food delivery application has been identified as a creator of fabricated content. The user published a post that gained significant traction, alleging that a company exploited its drivers and users. The post stated, "You guys always suspect the algorithms are rigged against you, but the reality is actually so much more depressing than the conspiracy theories."
Allegations and Viral Spread
The user claimed to be intoxicated and using public Wi-Fi to publish a detailed account of how the company allegedly used legal loopholes to withhold drivers' tips and wages. These claims resonated, partly due to a prior lawsuit against DoorDash for tip withholding, which resulted in a $16.75 million settlement. The Reddit post garnered over 87,000 upvotes and was cross-posted to other platforms, including X, where it received an additional 208,000 likes and 36.8 million impressions.
Investigation and AI Hoax Detection
Journalist Casey Newton of Platformer communicated with the Reddit poster, who then contacted Newton via Signal. The Redditor provided what appeared to be an UberEats employee badge photo and an 18-page "internal document." This document reportedly detailed the company's use of AI to determine a "desperation score" for individual drivers. Newton's attempts to verify the account led him to conclude that the information constituted an AI-generated hoax.
Newton noted that, historically, such a detailed document would have appeared highly credible due to the effort involved in its creation. However, the increasing sophistication of AI tools has altered the landscape of content verification.
Challenges in Fact-Checking with AI
Generative AI models can produce images and videos that are difficult to distinguish from authentic content, complicating fact-checking processes. In this instance, Newton utilized Google's Gemini, which identified the image as AI-generated due to Google's SynthID watermark. This watermark is designed to persist through common alterations like cropping, compression, and filtering.
Max Spero, founder of Pangram Labs, a company developing AI-generated text detection tools, commented on the issue. Spero stated that the volume of AI-generated content on the internet has increased, attributing this to the expanded use of large language models and other factors. He also mentioned instances where companies allegedly use AI-generated posts to achieve "organic engagement" and viral status on platforms like Reddit.
While tools like Pangram can assist in detecting AI-generated text, their reliability for multimedia content remains a challenge. Even when synthetic content is debunked, it may have already achieved viral spread. This development necessitates heightened scrutiny of content across social media platforms. It was also noted that multiple viral AI-generated hoaxes related to food delivery apps appeared on Reddit during the same period.