China & Generative AI

Lu Qi, the founder and CEO of Qiji Chuangtan, recently expressed his inability to keep up with the rapid pace of the big model era. He emphasized the importance of staying updated on papers and information and encouraged his subordinates to engage in "big model daily" activities. Lu Qi highlighted the challenges of managing the overwhelming amount of papers and codes in the field. He stressed the significance of the large model era and its impact on society. Lu Qi also shared his perspective on the Trinity Structure Evolution Model, which explains the interplay between information, model, and action systems in complex digital systems. He discussed how the transition from marginal to fixed costs in obtaining information has driven significant changes in society and industry. Lu Qi identified the large model as the core technology and basis of industrialization in the current inflection point. He emphasized that the large model revolution will impact all individuals, particularly those in the service economy, and underscored the importance of unique insights in the future. Overall, Lu Qi aimed to provide guidance to Chinese entrepreneurs, helping them understand the era of large models and find their position in it.

Lu Qi discusses three inflection points he sees in the future. The first inflection point is the ubiquity of models. He predicts that in the next 15-20 years, models will be integrated into various aspects of life and accessible through mobile phones and internet connections. The second inflection point is the widespread presence of automated and autonomous actions. The cost of physical actions will transition from marginal to fixed, leading to the integration of robotics and spatial computing. The third inflection point relates to the co-evolution of humans and digital technologies. OpenAI emphasizes the need for the emergence, agency, affordance, and embodiment of general intelligence. Lu Qi believes that OpenAI, with its groundbreaking work on large models, is positioned to lead in these inflection points and may surpass companies like Google in the future. He highlights OpenAI's key technologies and the importance of model architecture, specifically the Transformer model. He praises OpenAI's alignment engineering efforts to ensure models align with human values. Lu Qi acknowledges the challenges of developing large models, particularly the need for advanced infrastructure and the importance of tokens in language processing. He emphasizes that the large-scale encapsulation of knowledge within models is a significant breakthrough in natural language processing. The inflection point has arrived, and the paradigm shift is underway

In the future, models will be pervasive and integrated into various aspects of life. OpenAI's focus in the next 2-3 years is to make models more sparse, extend the attention window, and enhance recursion causality reasoning. They aim to improve model stability, token space, latent space alignment, and infrastructure tools. The influx of capital has set the growth flywheel in motion, with increasing investments and a growing business and profit model. However, OpenAI faces challenges in terms of social security and regulation, and efforts are being made to slow down the pace of advancement and ensure user feedback and risk mitigation. The development path revolves around the extensibility of models and the future model ecosystem. Lu Qi envisions more large models with complete world knowledge and improved learning, generalization, and alignment capabilities. Different types of models will be built, including domain-specific models and human models encompassing cognitive, task, and professional aspects. Lu Qi highlights the differences between human models and learned models, emphasizing the adaptability and scene-based nature of learned models while acknowledging the professionalism and limitations of human models. The future will see an ecosystem of models, enabling advancements in cognition and reasoning. Ultimately, grounding and the integration of perception and action are crucial for true intelligence. Lu Qi draws parallels between the model world and the biological world, viewing large models as genes that evolve and drive progress. Understanding the paradigm of this era, the focus shifts to embracing and leveraging the opportunities it presents.

The pace of development in the big model era is incredibly fast, with frequent "holy shit" moments.

• The cost of models has shifted from marginal to fixed, leading to structural changes and knowledge acquisition.

• Production capital has been improved, resulting in increased productivity and efficiency across industries.

• Large models will accelerate the development of robotics, automation, and self-driving cars.

• The widespread use of co-pilots is anticipated, with each individual potentially having multiple co-pilots to assist and automate tasks.

• Every industry will undergo structural impact and systematic restructuring.

• The formula of reducing costs and increasing efficiency will transform professions, making code farmers, doctors, and other professionals more efficient.

• Core industries like science, education, and healthcare will experience significant changes.

• Education will be influenced as ChatGPT-like models become resources for exams and defining the quality of a college student.

• The era is characterized by high-speed development and overwhelming structural changes.

The entrepreneurial opportunities in the big model era can be understood using a structured thinking framework.

• The foundation of digitalization includes platforms, open-source code, and open-source data, providing opportunities in areas such as digital applications and reducing costs.

• The three types of start-up companies are those focused on digital foundations, solving human needs through digitalization, and changing the physical world.

• Opportunities lie in carrying information, developing large models and their supporting infrastructure, exploring robots and autonomous systems, and carbon-based biological computing.

• Platform opportunities in the large model era are expected to be larger than information-based platforms, with characteristics such as working out of the box and having a simple and effective business model.

• The robustness and fragility of current models need to be addressed, and a balance between breadth and depth should be achieved in product development.

• Innovation should be driven by technology while meeting and understanding customer requirements in the implementation phase.

• The use of models like GPT-4 to develop GPT-5 amplifies the capabilities of coders, leading to different scale effects, competition patterns, intellectual property rights, and internationalization opportunities.

• China is seen as having significant opportunities in the big model era.