Written by: Techub News Compilation

Introduction

Recently, during a developer event hosted by AMD, a highly anticipated dialogue took place between two leading figures in the fields of AI and computing. On one side is Andrew Ng, a pioneer in deep learning and AI education and co-founder of Coursera; on the other side is Lisa Su, CEO of AMD, who has led the company's strong rise in AI hardware. This conversation was moderated by Sharon, head of AMD's developer ecosystem, focusing on the accessibility of AI, the value of an open ecosystem, and how developers can seize opportunities in this rapidly evolving era. In the context of increasing competition in AI infrastructure and the focus on ecosystem openness, the insights from these two leaders provide direction for the developer community.

Summary

• An open ecosystem is key to preventing "gatekeepers" from monopolizing and stimulating innovation at all levels of AI, especially in the semiconductor and foundation model layers.
• AI-assisted coding and modular tools are greatly lowering the barriers to application development, making "rapid prototyping - rapid failure - rapid learning" possible.
• AI will not replace developers but will create more demand for software; developers must continue to learn new skills to keep up with changes.
• The core advice for developers is: no need to wait for permission, boldly build with existing tools, and consider open options when choosing infrastructure.

Open Ecosystem: Counteracting Gatekeepers, Unleashing Innovation

At the beginning of the dialogue, Lisa Su emphasized AMD's core mission: to provide powerful and accessible computing roadmaps. She pointed out that the developer community is at the center of this ecosystem, and the recently launched developer cloud by AMD aims to lower the barriers to acquiring high-performance computing. She admitted that open-sourcing the software stack can sometimes be "scary" because imperfections will be exposed to the community, but it is precisely this openness and trust that allows the ecosystem to improve and advance faster through collective wisdom.

Andrew Ng approached the discussion from a macro perspective of the AI technology stack. He divided the AI stack into semiconductor layer (like AMD, NVIDIA), cloud hyperscale provider layer, foundation model layer (like OpenAI, Claude), and application layer. He believes that while technology and capital often focus on the underlying technologies, the application layer must create greater value since the revenue generated there will support the operation of the entire technology stack.

Ng particularly highlighted the importance of an open ecosystem in preventing the emergence of "gatekeepers." He used the mobile ecosystem as an example, pointing out that the two major platforms, Android and iOS, have, to some extent, restricted innovation and monetization methods in areas like keyboards, maps, and voice assistants. He worries that similar monopolistic forces could emerge at some layer of the AI stack, stifling innovation and extracting unreasonable profits.

"At the foundation model layer, supporting open-weight models is absolutely critical," Ng stated, criticizing certain companies for trying to lobby against open-source models. He believes that openness at this level has already generated tremendous vitality; for example, the DeepSeek model was quickly able to distill knowledge with the Qwen model soon after its release, and the community can rapidly expand the context window or add multimodal capabilities to the Llama model. These innovations are difficult to predict by a single company alone.

He further noted that there is also a potential risk of gatekeepers emerging at the semiconductor layer. Therefore, he highly praises AMD's efforts in promoting the open ROCm ecosystem, believing that ensuring developers have choices at the semiconductor level is key to preventing the rise of gatekeepers and allowing application developers to innovate freely. Lisa Su responded by stating that AMD is committed to simplifying upper-level development tasks to enable easy code portability, thereby giving developers true choices. She acknowledged the ROCm ecosystem is not perfect but is rapidly improving under the push from the open community.

Revolution in Development Paradigms: Rapid Prototyping and AI-Assisted Coding

When the topic shifted to how developers can engage, Andrew Ng painted a clear picture for developers at different levels. For the few developers working at the GPU kernel level, he advised deep involvement in the open ROCm ecosystem. However, for the vast majority of application layer developers, the ideal situation is to "not worry about the underlying layers at all" and just choose the optimal computing service after cost-performance calculations.

Ng highlighted two major transformations happening at the application layer. The first is the rise of AI building modules, such as prompt engineering, evaluation, RAG, vector databases, guardrails, and agent workflows. These modular components make it exceptionally fast to build complex applications, many of which were unimaginable just two or three years ago but can now be rapidly realized.

The second transformation is AI-assisted coding. Ng stated he cannot imagine going back to an era without AI-assisted coding. He believes that AI coding assistants are one of the fastest-growing directions in the AI field, from automatic completion in GitHub Copilot to more complex code generation and now to highly autonomous agent workflows, the evolution speed is astonishing. He mentioned that his team has recently used tools like Claude extensively; these highly autonomous AI coding assistants will greatly improve development efficiency and even help solve code portability issues between CUDA and ROCm.

The most direct impact of these tools is the explosive growth of "rapid prototyping" abilities. Ng shared that now he and his team can realize an idea in one afternoon, which might have required six engineers three months a few years ago. The iteration speed has also significantly increased: they can build a prototype in a day, get user feedback the same evening, and start the next round of iteration the following morning.

"My focus is more on bringing the cost of proof of concept down to extremely low levels," Ng explained his methodology, "so that you can build 20 prototypes, and even if 18 of them fail quickly, finding two that are truly valuable is a very profitable deal." He believes that whether for startups or large enterprises, innovation processes are changing, and utilizing this "crazy fast" prototyping capacity to explore and discover genuinely feasible solutions.

Lisa Su fully agreed and connected the concept of "rapid failure, rapid learning" to AMD's developer ecosystem strategy. She admitted that a year ago, AMD might have been more focused on large-scale customers, but now recognizes the incredible speed and density of innovation in the startup community. Therefore, AMD has adjusted its developer strategy to shift towards a more agile release rhythm, such as launching new training and inference Docker images every two weeks to quickly collect feedback and iterate. She disclosed that AMD has even started training AI to write kernel code to accelerate the optimization process.

Developers in the AI Era: More Demand, Higher Expectations

In response to the widespread anxiety of "Will AI replace programmers?", Andrew Ng provided a clear and forceful rebuttal. He referred to the earlier argument that "AI will automate coding, so there is no need to learn programming" as "one of the worst career advice ever." His logic is that when something becomes easier, more people should do it, not fewer. History has shown that with every reduction in programming barriers, from punch cards to keyboards, from assembly language to high-level languages, there has been a surge in the number of programmers and a flourishing software industry.

Ng believes that the demand for customized software has almost no practical ceiling. As AI-assisted coding reduces development costs, the world will produce more and greater software. However, the nature of software engineering is transforming. He observed that recent graduates who are adept at using AI tools may have productivity far exceeding that of seasoned engineers who still rely on the work methods established before the explosion of generative AI in 2022.

But this does not mean experience is no longer important. Ng pointed out that the best engineers he knows are those who possess a deep understanding of system architecture and computer science, coupled with proficient AI skills. The current market contradiction lies in that many companies cannot find enough developers skilled in generative AI applications, while many educational systems (especially university courses) are updating slowly and failing to train students to meet these urgent needs.

Lisa Su provided corroboration from the corporate perspective. She stated that AMD is hiring extensively and utilizing AI throughout the entire process, including hardware design, software development, sales, marketing, and human resources. The goal of AI is not to replace jobs, but to make employees more capable and efficient, thereby accelerating product launch cycles and expanding project pipelines. She emphasized that continuous learning is key, and people need to constantly update their knowledge through work, courses (such as Ng’s courses on Coursera), and exchanges like this forum.

Action Guide for Developers: Build Boldly Without Permission

At the end of the dialogue, the host asked both guests to give the on-site developers a "call to action." Andrew Ng shared one of the most important life lessons he believes: "You are allowed to take direct action." As long as it does not harm others or cause destruction, there is no need to wait for others’ permission or for them to agree with your ideas; just build.

He pointed out that, supported by AI-assisted coding and powerful building modules, the number of things individuals can achieve has exploded. This is a fantastic time to build. Even if what is constructed does not meet expectations, the process itself is a valuable learning opportunity. He reminisced that Coursera also started as an "amateur project," and many great projects in Silicon Valley originated when someone took action to solve a problem without seeking permission (but still responsibly).

Lisa Su's call was even more direct: "Build, build, build, and build on AMD." She hopes developers will give AMD's ROCm ecosystem a chance and reiterated that developers are at the center of all AMD's efforts. She compared the current stage of AI development to "the race has just begun," and over the next five to ten years, there will be progress beyond imagination, with the developer community having the opportunity to jointly write the future's chapters. AMD hopes to be a partner and enabler for developers, learning and growing together with them.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。