Key Takeaways from Zhou Hongyi’s Speech at the 27th Harvard China Forum on Large Model Opportunities in China

On April 13 (Eastern Time), the 27th Harvard China Forum was held at Harvard University in Boston, USA, where Zhou Hongyi, founder of 360 Group, delivered a speech. He said developing specialized large models could be a huge opportunity for China, and all enterprises should let specialized large models reshape themselves. He invited Harvard graduates to return to China for large model entrepreneurship, jointly forging China's AI path.

1. Artificial intelligence will not necessarily eliminate individuals; those who do not understand AI will be eliminated by peers who use AI.

2. It's not only OpenAI's path; OpenAI believes scaling law, more data, larger models, more compute. Models keep getting bigger and will eventually break through to super strong intelligence. China is also on this path, but it takes time.

3. Large models are not only about building super general models; besides getting bigger, they can also get smaller. Large models need to be on terminals, computers, phones, cars.

4. Computers entered millions of homes and industries, sparking the real industrial revolution. Therefore, we must bring large models down from the altar and truly integrate them with many industries so enterprises can actually use them.

5. Enterprises need to find vertical scenarios and train specialized large models; in the future, enterprises and governments will not have only one general model.

6. In the future, enterprises will have dozens of specialized large models each solving specific problems, cultivating specialized experts rather than a jack‑of‑all‑trades.

7. Open source models are not yet at GPT‑4 level, but focusing on a direction and strengthening with proprietary data can surpass GPT‑4 in specialized capabilities.

8. Building specialized large models does not require trillion‑parameter models; tens or hundreds of billions of parameters can run on consumer‑grade GPUs, reducing costs from hundreds of millions to tens of thousands or a few hundred thousand RMB, making them affordable for many enterprises.

9. 360 has built a security large model that far exceeds GPT‑4 in specialized security offense‑defense knowledge and automatic attack detection.

10. Many entrepreneurs first build a large model then look for scenarios, which is wrong; users want solutions to problems, not just a model.

11. Large‑model entrepreneurship must first find star, pain‑point, or must‑have scenarios, define functions based on them, then train specialized models.

12. A pain‑point must‑have scenario is one where AI can boost efficiency or experience tenfold or cut cost and labor tenfold in internal business processes or external product/service functions.

13. For many entrepreneurs, product and model are integrated: first have a scenario, define product, then let product define its own proprietary model. Building enterprise‑grade specialized large models offers huge entrepreneurial opportunities.

14. China may lead a new industrial revolution. China has the most traditional industry categories, complete productivity and supply chains, all undergoing digital‑intelligent transformation. All enterprises need specialized large models to reshape everything; all industrial and enterprise software should be rewritten.

15. AI+ today is essentially large‑model+, not just building a model but truly empowering its capabilities across traditional industries, with broad development potential.

16. If you have good large‑model technology, you are welcome to return to China to start a business and cooperate with us.