The Grok4 launch event can only be described as mediocre, and the specific performance is a bit unsatisfactory after trying it Here's a summary of 👇 the presentation 1. Performance: Grok 4 performs well in multi-domain tests, with a far higher accuracy rate than similar models in difficult questions such as the "Ultimate Human Exam", a better multi-agent version, and full scores in programming, mathematics and other tests, and its academic ability has reached the graduate level, surpassing most humans. 2. Training development: From Grok 2 to 4, the amount of training increases by orders of magnitude, and the 4th generation is 100 times higher than the 2nd generation. Generations 3 to 4 focus on reasoning and reinforcement learning, with the help of data augmentation and other technologies and supercomputers to achieve principle reasoning and self-correction. 3. Functional application: The delay of voice interaction is halved, and natural voice is added; After the API is opened, it is widely used in business simulation, scientific research, game development and other fields, such as increasing the net value of vending machine business, accelerating scientific research, and rapid game development. 4. Future plans: Launch coding models in a few weeks to improve multimodal capabilities; In the next 3-4 weeks, video generation training will be launched, with the goal of building faster and smarter models and promoting the upgrading of human civilization.