What is what The future of the same button in the era of Artificial intelligence? Max levchinThe co -founder of Paypal and claims the CEO, has a new and extremely valuable role to prefer data to train AI to make a more appropriate conclusion for human decision makers will make.
It is a famous difficulty in learning that a computer with a clear reward function will participate in continuous consolidation learning to improve its performance and maximize that reward but this optimization path often leads AI systems to very different results compared to the results of human judgments.
To introduce a remedial force, AI developers often use what is called enhanced learning from human feedback (RLHF). Basically, they are putting a human thumb on the scale when the computer reaches its model by training it on data that reflects the practical hobby of people. But where does human preferences come from and how many of which are needed for input? So far, this is a problem for RLHF: It is a costly method if it requires hiring supervisors and human captions to enter feedback.
And this is the problem that Levchin thinks that it can be solved by the same button. He sees the accumulated resources that are now in FacebookIs a god for any developer who wants to train a smart agent about human preferences. And how a big deal? I will argue that one of the most valuable things Facebook owns is the preferred data mountain, Lev Levchin told us. Indeed, at the bending point in the development of artificial intelligence, have access to what the content is, used to train AI models, perhaps one of the most valuable things on the Internet.
While Levchin imagine who learns from human preferences through the same button, who has changed the way these hobbies are shaped from the beginning. In fact, social media platforms are actively using AI not only to analyze likes, but to predict them, the ability to display the button itself is outdated.
This is a prominent observation for us because, as we talked to most people, the predictions mainly from a different angle, describing not the same button that will affect whose performance but who will change the world of the button. We have heard, who is being applied to improve social media algorithms. In early 2024, for example, Facebook tested Use AI To redesign the video recording algorithm for users. Can it give a better weight of the variables to predict which video that users want to watch the next? The results of this initial test showed that it could: applied AI to the task paid during longer tracking, Facebook performance data hoped to strengthen.
When we ask YouTube The co -founder of Steve Chen what the future keeps the same button, he said, sometimes I wonder if the same button is necessary when anyone is sophisticated enough to tell the algorithm with an accuracy of 100 percent that you want in a simple way.
However, he continues to point out that a reason that the same button may always be necessary to handle sharp or temporary changes in the needs of viewing for events or situations of life. There are days when I want to see the content a bit more suitable, say, my children, he said. Chen also explains that the same button may have a life expectancy because its role in attracting advertisers, other important groups with viewers and creators because of the similar activities as the simplest hinge can connect those three groups. With a hose, a viewer simultaneously conveys the appreciation and direct response to the content provider and evidence of the participation and priority to the advertiser.