OpenAI's AI can now operate computers like humans

OpenAI is getting closer to its original goal: making AI agents operate computers as smoothly as humans, especially when using browsers. According to Business Insider, the company has started giving some customers access to its computer-operating feature and plans to expand it. The feature relies on data and training methods focused on improving computer skills, including human demonstrations and reinforcement learning, so the model can do tasks like filling out tax forms or creating 3D models.

Now, ChatGPT can read a webpage's underlying information directly after opening it, reducing the need for constant screenshots, and it can use screen recording to quickly understand what's on display. Ali Weinstein, who leads the computer-operating team, says that once AI operates computers faster than people, it will change how we interact with them. However, Yu Zhou, a researcher at Columbia University, notes that the speed still falls short of what regular users expect, especially when it comes to adapting to any website or app. OpenAI is also working on safety measures, like asking for user permission before sending data or deleting content, to balance ease of use with security.