
链研社|Jul 10, 2025 09:19
The Grok4 press conference can only be described as mediocre, but after trying it out, the specific performance was somewhat unsatisfactory
The following is a summary of the content of the press conference 👇
1. Performance: Grok 4 performs excellently in multi domain testing, with accuracy rates for difficult problems such as the "Ultimate Human Exam" far exceeding similar models. The multi-agent version is even better, and it has achieved full marks in programming, mathematics, and other tests. Its academic ability is at the level of graduate students, surpassing most humans.
2. Training development: From Grok 2 to 4, the amount of training increases by orders of magnitude, with 4 generations being 100 times more than 2 generations. Generations 3 to 4 focus on reasoning and reinforcement learning, utilizing data augmentation techniques and supercomputers to achieve principle reasoning and self correction.
3. Function application: Half the delay of voice interaction, add natural voice; After the API was opened, it has been widely used in fields such as business simulation, scientific research, and game development, such as increasing the net value of vending machine business, accelerating scientific research, and rapidly developing games.
4. Future plan: Launch coding models within a few weeks to enhance multimodal capabilities; In the next 3-4 weeks, video generation training will be launched with the goal of creating faster and more intelligent models to promote the upgrading of human civilization.
Share To
HotFlash
APP
X
Telegram
CopyLink