In the latest episode of "Walking with Bill Gates: Eliminating Confusion," Sam Altman, CEO of OpenAI, pointed out that OpenAI is creating newer and better models on this long and continuous curve. He emphasized the key feature of multimodality in GPT-5, which enables it to handle video inputs and generate new videos. Altman believes that in the near future, AI will be able to handle more complex tasks, thus improving productivity. "You can imagine telling a little assistant, 'Go write this whole program for me.' It might ask you some questions, but it's not just writing a few functions—it will open up a lot of new things," Altman said. He jokingly said, "Maybe one day there will be an AI that you can say, 'Start and run this company for me.' And then one day, there might even be an AI that you can say, 'Discover new physics [laws] for me.'" Currently, several foundational models have achieved the capabilities of GPT-4. Google may soon release Gemini Ultra. Meanwhile, Arthur Mensch, CEO of Mistral, announced on French National Radio that the company will launch an open-source GPT-4 level model in 2024. "Will Gemini Ultra be released before GPT-5? I doubt Gemini Ultra surpasses GPT-4 in reasoning," wrote Bindu Reddy, CEO of Abacus AI, in a post on X. It is worth noting that Mistral Medium has received over 6,000 votes and performed exceptionally well, reaching the level of Claude. "Mixtral will surpass GPT-4 this year. Today, in the chatbot arena, it is the only top-tier open-source model (following GPT-4, Claude, and Mistral Medium), and it is the smallest one with 7 billion instructions. It is even better than Google's Gemini Pro," said AI expert Santiago with delight. Facts and statements like these make it necessary for OpenAI to remain vigilant. What's next—GPT-5? In the conversation with Gates, Altman discussed GPT-5 in detail, emphasizing customization and personalization. "Understanding you, your emails, your schedule, how you like to book appointments, and connecting to other external data sources. These will be the most important areas of improvement," Altman said. Furthermore, he claimed that GPT-5 will have better reasoning capabilities than GPT-4. "GPT-4 can only reason in extremely limited ways. Additionally, reliability is also an issue. If you have GPT-4 answer most questions 10,000 times, one of those times might be pretty good, but it doesn't always know which one. But you want the best response out of those 10,000 every time," Altman said. In addition to new models, OpenAI may soon start focusing on specific verticals to serve its customers. "Programming is probably the most obvious area where we can improve productivity right now, and AI has been massively deployed and used in this field. Healthcare and education are also on this curve, and we are excited about them too," Altman said. Meanwhile, OpenAI recently launched the GPT Store. This may help OpenAI develop GPT-5 as they will obtain large-scale labeled data from GPTs created by customers using personal data. Interestingly, the blog did not mention that OpenAI will not use GPT Store data to train its models. OpenAI is about action The OpenAI team is very active on X, seeking customer feedback to improve its models. Recently, Greg Brockman, President of OpenAI, asked a question on X: "How has ChatGPT changed your life?" A user named Aaron Stormerr shared the profound impact ChatGPT had on their academic journey as a blind computer science student. Similarly, Logan Kilpatrick, Developer Relations Lead at OpenAI, asked, "Who is building the most useful/coolest product using the OpenAI API? I want to spend more time in 2024 listening to feedback from builders about how we can support them/you." Clearly, OpenAI's approach is to release an imperfect product or feature and then improve it through continuous consumer feedback, launching products as quickly as possible. Therefore, the company's success depends on the next model, similar to what Walt Disney once said to his team, "Our success and failure hinge on our next picture."