DeepSeek has created a huge wave of discussion and panic in the market.
I think it’s great overall.
- Innovation
From a technology standpoint, DeepSeek is truly pushing the envelope in both training methods and model architecture. They’ve combined many good ideas from industry and academia and, more importantly enhanced them to create solutions that are both more capable and more cost-effective. Their work has single-handedly revived the MoE (Mixture of Experts) architecture, which made quite a splash some time ago but had since grown quiet. These advancements will undoubtedly give AI researchers a lot to think about.
- Open-Sourcing
Not only have they open-sourced their model architecture and weights, but they’ve also dived into extensive details about their ideas, thinking processes, improvements and even their failures in their technical reports. This level of transparency outshines what we’ve seen from OpenAI, Meta, Google, and xAI, and it significantly advances the open-source community. In contrast, all the major tech companies mentioned above have been notably guarded about sharing their methods and learnings.
- Rethinking
Their success offers a different perspective for the industry—particularly when it comes to re-examining the scaling laws and the hyped AI infrastructure spending. Deflating some of the bubble will encourage more sustainable and healthier growth. Instead of endlessly chasing bigger datasets and more compute, we should consider efficiency. In retrospection, there are still low-hanging fruits waiting to be harvested.
- Competition
Competition is always beneficial for technological advancement.
- Skepticism
There’s been a fair amount of skepticism around DeepSeek’s cost claims; I share some of that uncertainty. However, because everything is open-sourced, the community can, and already has begun to, reproduce their work. This means we’ll soon learn the truth. Even if their claims of achieving such an amazing results with limited resources prove accurate, the accomplishment might not be as groundbreaking as many layman think. Following the path of pioneers is usually less challenging than blazing the trail yourself.
- Censorship Concerns
Some other critics question how successful DeepSeek will be with its built-in censorship. Personally, I’m not too worried. The Chinese market alone is enormous. And I naively hope their global success might eventually lead to looser regulations and censorship from the Chinese government.