Recently, Robin Li made critical remarks about open source large models in an internal speech, arguing that open source large models are of little significance. This has sparked a debate in the industry about the open source vs. closed source debate.
Robin Li believes that open-source large models are often only verified in small-scale scenarios, and have not been verified by large-scale commercial verification. And open source will not achieve a high flame for everyone to pick up firewood, the key is to rely on the investment of large companies.
In contrast, closed-source models have business models that can gather talent and computing power, and ultimately create more powerful models.
However, there are also opinions that open source is conducive to lowering the barrier to use and allowing more developers to participate in innovation. Faster progress can be achieved through open source community collaboration. In addition, open source is beneficial for improving the security and explainability of models.
In my opinion, open source and closed source have their pros and cons, and should not be simply opposed. At present, open source and closed source form a complementary development trend.
On the one hand, the industry's top large models are indeed closed-source, such as OpenAI's GPT series, Baidu's Wenxin series, etc.
These closed-source large models have more high-quality data, computing power, and talent support, with a large number of parameters and strong performance. Closing the source code also helps the company control the core technology and achieve commercial operation.
On the other hand, open source communities are also rising rapidly, such as Claude of Anthropic. The open source model promotes technological innovation and industrial development in another way.
It lowers the barrier to entry, and small businesses can also take advantage of the open source big model by fine-tuning. Open source code is beneficial for the community to work together to improve the model and identify problems. The open-source model supports developers to innovate based on the model, which will give rise to more downstream applications.
In short, closed-source emphasizes independent R&D and commercialization, while open source emphasizes shared and collaborative applications. When it comes to entrepreneurs, Robin Li believes that there should be no overemphasis on developing their own models, but rather on application innovation. I agree with that.
Start-up companies have limited resources, so they should choose key breakthroughs. It is not wise to compete directly with the giants in the model, and the model should be leveraged in a vertical to provide unique value.
In general, open source and closed source should coexist and co-prosper to achieve positive interaction. When it comes to entrepreneurs, they should still focus on business innovation rather than pure technological innovation.
In the age of AI, we need an open and collaborative attitude. Whether it's open source or closed source, the goal should be to drive progress across the industry. At the same time, it is also necessary to be rational and not to be extremely demanding or judgmental about a certain way.
Only by embracing diversity can we unleash the vitality of innovation and enable AI to benefit more fields.