Wu Enda on AI Modeling Strategy: Technology Selection and Values Consideration from DeepSeek, Gemini

Recently, the field of large-scale language modeling has taken off.Google (used form a nominal expression) Gemini Continuous iteration of the series model ( Google Releases Gemini 2.5: Big Improvements in "Thinking" Capabilities ), and from China DeepSeek A new version of V3 has been released ( DeepSeek-V3 Model Low-Profile Update, Code Capability Jumps to Claude-3.7 ), intensifying competition in the basic modeling market. Companies such as Baidu are also actively developing models that can compete with OpenAI The model of resistance ( Baidu Releases Wenxin Big Model 4.5 and X1: Dual Evolution of Multimodal Capabilities and Deep Thinking This signifies that the global competition for AI-based models is no longer just a U.S. stage.) This signifies that the global competition for AI-based models is no longer just a U.S. arena, and that Chinese power is accelerating its entry into the game.

In this context, renowned AI scholars,AI Fund Managing Partner and DeepLearning.AI creator Andrew Ng(Wu Enda) shared his insights into the current AI landscape during a recent appearance at a tech event.Andrew Ng past experience in Google Brain cap (a poem) Baidu He holds key AI leadership positions and his perspective is uniquely valuable in understanding the AI landscape in the US and China.

Companies should adopt a flexible multi-model strategy

faced with Llama,DeepSeek,通义千问 (Qwen) and many other models have emerged.Andrew Ng noted that Open Weight Models (OWMs) are becoming a key component of the AI supply chain. He believes that these top-performing models, whether from the U.S. or China, are reshaping the global digital technology landscape.

From an enterprise application perspective, the intense modeling competition has brought obvious benefits - the cost of model use continues to fall, driving accelerated innovation at the application layer.Andrew Ng shared his team's practical experience: the core strategy is not to bind to a single model vendor, but to build a flexible technical architecture so that the most suitable model can be switched to at any time according to task requirements, cost-effectiveness and performance. He revealed that his team is currently adopting the strategy of multiple models in parallel.

(go ahead and do it) without hesitating DeepSeek and other models have received attention for their performance and openness, but some organizations are still hesitant to adopt their APIs due to data security and compliance concerns. however.Andrew Ng It is argued that, in addition to these obvious factors, there are deeper considerations.

Ng notes that while the likes of DeepSeek These types of open weighting models often conjure up images of the Chinese companies behind them, but there's no denying that the role of such models in the AI supply chain is becoming more and more critical.

Values and geopolitical considerations behind model selection

Andrew Ng As a reminder, when a business or individual user interacts with AI models for an extended period of time, a larger question needs to be pondered, "Do these models reflect the values of the country or business in which they are published?"

AI models are not the product of a technological vacuum. Through conversations, content generation, and even casual chats, users may be subconsciously exposed to and influenced by the worldview embedded in the model's training data. This is reflected in word preferences, interpretations of specific legislation, and may even touch on attitudes toward sensitive topics. When a user asks about culturally relevant or controversial topics, the model's response may indirectly or directly convey the position of the developing country or company.

Ng explained that when people around the world use AI conversational services and ask about borders, cultures and sensitive topics, the country or company that develops the model has an impact, either directly or indirectly.

This is not only a challenge that companies need to face when making localized applications, but it may also have a long-term impact on the conceptual system of the whole society. This explains why there are calls in some regions for the development of localized language models aimed at preserving local cultural characteristics and meeting the needs of specific business scenarios.

Andrew Ng Affirmative. DeepSeek and other Chinese models have contributed to the technology community, and notes that both Chinese and American companies are adopting these models. But he also raises a key question, "Can other countries and regions also invest enough resources to sustainably compete in open weighting models?" He argued that openness accelerates knowledge dissemination, and while it may benefit competitors, it often ends up benefiting the initiating country the most. When a country's open model is widely used, that country will undoubtedly gain significant influence.

Open weighting models: resisting monopoly and accelerating the dynamics of innovation

From another perspective, the existence of open weighting models (usually meaning that the weights are publicly available, but the training data and methods may not be fully open source) is crucial to prevent market monopolization.Andrew Ng Adding that in the absence of such models, numerous companies may be forced to rely on a handful of tech giants that hold powerful arithmetic resources, thus increasing market concentration.

Currently, the open camp (which includes open weights and fully open source models) and the closed source models (such as the OpenAI (used form a nominal expression) GPT-4The competition between the two is becoming more intense by the day. While the open camp is still playing catch-up in some areas, the potential it shows can no longer be ignored. Regardless of how the competitive landscape evolves, businesses and developers around the world will be the beneficiaries.

Seize the moment and build applications with AI

Andrew Ng Ultimately, it sends a clear signal to businesses and developers everywhere that utilizing the AI The time is ripe to build services and drive innovation. He emphasized that the tech giants have invested heavily in developing advanced tools that make it easier than ever for anyone to develop AI applications. He encouraged applying these capabilities across industries such as semiconductors, manufacturing, healthcare, and more, "There's no better time to act than now."