When OpenAI allowed users to create images with ChatGPT, some users exploited it to create fake receipts. Some worry that fraudulent crimes could occur, such as receiving refunds with fake receipts.
Didi Das, a venture capitalist active on social media, recently posted a fake receipt photo of a San Francisco steakhouse made from GPT 4o on X, according to TechCrunch and other foreign media.
Other users then made fake receipts and added food or drink stains to make them look more realistic. One user even made a fake wrinkled receipt for a local chain of restaurants in France and posted it on social media.
Fake receipts made from GPT 4o have several flaws. Commas and periods appear mixed, and basically, the sum of amounts does not match. This is because LLM (large language model) is having difficulty with basic computations.
Some point out that it will not be difficult for those who want to commit fraudulent crimes by exploiting GPT 4o to compensate for their laxity with photo editing programs or more accurate prompts. Some observers say that if the generation of fake receipts becomes easier, there will be cases where fake expenses will be refunded.
An OpenAI official said, “All images contain metadata indicating that ChatGPT produced,” adding, “If users violate the policy, we will take action and the company is supplementing it with feedback.” He added, “The creation of fake receipt images gives as many people as possible freedom to create in non-crime situations such as original art or advertisement production.”
SALLY LEE
US ASIA JOURNAL
Arabic
Chinese (Simplified)
Dutch
English
French
German
Italian
Japanese
Korean
Portuguese
Russian
Spanish