Видео
IMG_8361.MP4 · 168.0 MB · click to show
IMG_8361.MP4 · 168.0 MB · click to show
🗣Kimi CEO: "AI is Accelerating Faster Than Ever. A new Scaling Era has Begun."
- AI is improving in two dimensions, vertical scaling and horizontal scaling.
- Reasoning has advanced dramatically. AI now solves complex math and coding tasks that were nearly impossible just a year ago.
- Context windows have exploded from roughly 4K to 8K tokens to 128K+ tokens, allowing models to understand far longer documents and conversations.
- Progress is no longer driven only by larger models. Better post-training, reinforcement learning, higher-quality data and inference-time reasoning are now accelerating capabilities.
- Multimodal AI is advancing rapidly. Systems can now generate videos, transform research papers into realistic podcasts and seamlessly work across text, images, audio and video.
- OpenAI's o1 introduced a new paradigm. Experts described it as AI demonstrating System 2 reasoning, enabling deliberate thinking, self-correction, exploration of multiple solution paths, and reflection.
- A new scaling law has emerged. Beyond pre-training, reinforcement learning and inference-time search now provide another predictable path for improving intelligence.
- AGI has entered a new stage. According to Professor Ju Jun's framework, AI has moved from L1 Chatbots toward L2 Reasoners, representing a major qualitative leap.
- Zhilin Yang (CEO of Kimi): "The most important significance is that it raises the upper limit of what AI can achieve."
- Professor Ju Jun: "The progress curve is getting steeper and steeper."
- Industry experts: "We're in the middle of an acceleration process, and the pace of AI development is still increasing."
@aipost 🏴
8.7K ·