AI model Sora surprises the world

Meet Sora, the cool model from OpenAI. It can turn your words into videos like magic (魔法)! Just tell Sora your idea, and watch it make a video right before your eyes.

How good is it?

Videos made by Sora look good on both computers and smartphones. Sora can also make videos with lots of people doing different things with clear details (细节). It also knows how things should look in real life.

How does work?

Sora looks at many videos and learns from them by breaking them into very small pieces. It uses these pieces to make a new video. The new video is gray at first and very messy (乱糟糟的) —you can’t tell what’s in it. Then, Sora fixes the video until it looks good and smooth.

Sora also gets help from ChatGPT. It turns the user’s short sentences into clear instructions (指令). This helps Sora make the video just as you asked.

Is it good enough?

Sora may find “busy” pictures hard to understand. For example, when a person moves, things around them change (变化). Sora may not be able to tell “left” from “right” and mess up a person’s leg moves.

Robot fools humans

The idea that robots could one day be our friends has been around for a long time but until now they have always seemed more machine than person. A new study by a researcher at the Italian Institute of Technology has found that in fact it’s possible for robots to fool us here and there, they are just less perfect than human. Inspired by the Turing Test, Agnieszka Wykowska came up with an experiment using a child-sized human-like robot called iCub. Sitting in front of a screen, the people in the test were asked to press a button when a green box appeared on the screen, however, iCub only responded to red boxes Wykowska programmed the robot to take different lengths of time to respond, just like a person would. Afterwards, the humans were asked to judge whether they had been competing with a machine or a person according to the time of the buttions being pressed—and they couldn’t tell. Wykowska believes that making robots have more human responses is the key to making better robots and creating more trust between humans and robots.

In another study with iCub, the robot and a human watched special videos of animals. In the first set of tests, iCub greeted the person, asked for his name and kept eye contact with its camera “eyes”. It made comments about the videos, laughed and showed emotions (情感). In the second experiment, however, iCub did not communicate with human and just made machine-like noises. The researchers found that people believe that the more human iCub is, the more it has its own thoughts and desires. The team now wants to know if people can communicate with robots that do not look human but that still show human-like actions. The researchers plan to repeat the experiments, using robots of different shapes and sizes, to see how people give their responses.