Tech

People Trying to Use Facebook's Leaked AI to Improve Their Tinder Matches

The results aren't as impressive as GPT4, but the news shows what sort of things people will use leaked AIs for.
Tinder
Image: NurPhoto/Contributor
Screen Shot 2021-02-24 at 3
Hacking. Disinformation. Surveillance. CYBER is Motherboard's podcast and reporting on the dark underbelly of the internet.

Users of Facebook’s leaked artificial intelligence are tasking the tool with generating text for their Tinder profiles and things to say during conversations in the hope of getting a real world date.

Although it’s unclear if participants have had any tangible success yet, it still demonstrates how Facebook’s LLaMA model is being used in the wild after the company lost control of it in a leak earlier this month. Last week, OpenAI released GPT4, its latest iteration of its own AI. That comes with many guardrails and stops users from asking the AI to generate all sorts of material. In Motherboard’s own tests, GPT4 will generate text for Tinder profiles if asked to do so, and do a better job than current implementations of Facebook’s leaked model. Still, Facebook’s leaked model has no protections in place, meaning users are free to experiment with it however they see fit, with AI-dating likely just being one of the first applications. 

Advertisement

“The plan is to develop a series of prompts so that the AI 'assist' you in establishing a conversation. I believe this can encourage many people to start chatting,” Alfredo Ortega, an information security software engineer who created a Discord bot for others to interface with Facebook’s leaked AI, told Motherboard in an online chat.

Do you know anything else about the LLaMa leak? Are you using it for any projects? We'd love to hear from you. Using a non-work phone or computer, you can contact Joseph Cox securely on Signal on +44 20 8133 5190, Wickr on josephcox, or email joseph.cox@vice.com.

On his Discord server, Ortega laid out the process for using the bot to generate material for Tinder. “The aim of this channel is to conduct continuous research to discover the smallest LLM model capable of arranging a date with a human, male or female. Use the AI as a proxy: Give the AI the conversations with your match, and write back the answer to the dating-app chat. Please anonymize any information before posting,” Ortega wrote.

Earlier this month Ortega used Facebook’s leaked large language mode to make the Discord chat bot which he called “BasedGPT.” Elon Musk previously said he wanted to create a “based AI” as an alternative to Bing’s ChatGPT, which he and other conservatives believe is too “woke.”

Advertisement

Facebook originally made LLaMA only available to approved researchers and other parties, but a 4chan user quickly leaked it on the forum.

One user in the Tinder-focused channel wrote “I've got 50 likes on Tinder because I never actually swipe right on anyone, I just watch the number go up as a vacuous sense of validation. This will finally allow me to put those likes to use, lmao.”

The Facebook AI-powered bot is fairly basic, but it is able to generate what a conversation on Tinder may look like and complete the text of potential Tinder profiles when given a prompt. In one example, the AI generated messages such as “Hey, I like your dog. Wanna get a drink sometime?” and “There’s this new bar that just opened up on Main Street. Have you been there yet?”

“Hmm, promising,” one user wrote in response. The AI did have a problem generating a “flirty” conversation, though.

At the moment, the Tinder project requires a user to manually take the bot’s output and paste it into Tinder themselves if they wish. Ortega said it’s “just a matter of time” until someone automates the process.

“Conversations in online-dating are not very complex, mostly about setting up a date, I believe even a very small LLM [large language models] can automate this,” Ortega told Motherboard.

Advertisement

Facebook previously told Motherboard in a statement that “We have made clear that the LLaMA foundational models were released by Meta for the purposes of research only. In line with industry practices, if we find a suspected violation of the LLaMA research license we investigate.”

Tinder did not respond to a request for comment on whether using AI to generate parts of profiles or conversations is against its terms of use.

Last week Motherboard reported on CupidBot, an app allegedly made by disenfranchised ex-Tinder employees which uses AI chatbots to speak to women on men’s behalf. The purpose is to combat the “disadvantages the average man” faces in online dating. (Later, the creators retracted those claims about being ex-Tinder employees, and told Motherboard “We do not and have not hired any ex-Tinder engineers.”)

Update: This piece has been updated to include more information from the creators of CupidBot.

Subscribe to our cybersecurity podcast, CYBER. Subscribe to our new Twitch channel.