What is The future of similar buttons in the era of artificial intelligence? Paypal co-founder and confirmed CEO Max Levchin saw new roles in liking data to train AI to be more consistent with the conclusions of those human decision makers can come up with.

This is a well-known puzzle in machine learning. Computers with explicit reward capabilities will participate in relentless reinforcement learning to improve their performance and maximize rewards, but this optimization path often leads to a very different outcome of AI systems than those caused by human exercise of human judgment.

To introduce corrective power, AI developers often use reinforcement learning from human feedback (RLHF). Essentially, when computers train data to reflect the actual preferences of data, they put the human thumb on scale. But where does this human preference data come from and how much does it take to input to be valid? So far, this has been a problem with RLHF: It is an expensive approach if it requires hiring human supervisors and commenters to enter feedback.

This is a problem that Levkin believes can be solved with a similar button. He believes that today, today, the accumulated resources in Facebook’s hands, as any developer who wants to train smart agents on human preference data. How big is that? "I think one of the most valuable things Facebook has is those mountains that love data," Levkin told us. Indeed, at this point in the evolution of artificial intelligence, access to "what humans like, used to train AI models, is probably one of the most valuable things on the internet."

Although Levchin imagines AI learning from human preferences through similar buttons, AI has changed the shape of these preferences. In fact, social media platforms are actively using AI, not only to analyze likes, but also to predict them, which can make the button itself obsolete.

This is a stunning observation for us, because when we talk to most people, these predictions come from a different angle and do not describe how similar buttons affect AI's performance, but how AI will change the world of buttons. We have heard that AI is being applied to improve social media algorithms. For example, in early 2024, Facebook tried to use AI to redesign an algorithm that recommends reel videos to users. Can we propose better variable weighting to predict the video users want to see the most? Results from this early test suggest that it can: apply AI to tasks with longer viewing times – performance metrics Facebook hopes to improve.

When we asked YouTube co-founder Steve Chen how the future of the button press is, he said, “I sometimes wonder if you want to simplify as you watch and share when AI is complex enough to tell the algorithm with an algorithm that is 100% accurate, but the goal to simplify is, however, the ultimate ability of the button is ultimate, but the accuracy of the button is ultimate, but it is ultimate, but it is ultimate, but it is simplified, but it is simplified as it can be made available.”

He continued to point out that, however, the reason why buttons are always needed due to life events or circumstances may always be the reason for a sharp change or temporary change in viewing needs. "Sometimes, what I want to see is more relevant to my kids," he said. Chen also explained that similar buttons may have longevity because of their role in attracting advertisers (with another key group next to audiences and creators), because similar is the simplest hinge that connects these three groups. A Tap audience also conveys appreciation and feedback directly to content providers, as well as evidence of participation and preference for advertisers.