Andrew Ng

I recently spoke at the Sundance Film Festival on a panel about AI. Sundance is an annual gathering of filmmakers and movie buffs that serves as the premier showcase for independent films in the United States. Knowing that many people in Hollywood are extremely uncomfortable about AI, I decided to immerse myself for a day in this community to learn about their anxieties and build bridges. I’m grateful to Daniel Dae Kim @danieldaekim, an actor/producer/director I’ve come to respect deeply for his artistic and social work, for organizing the panel, which also included Daniel, Dan Kwan, Jonathan Wang, and Janet Yang. I found myself surrounded by award-winning filmmakers and definitely felt like the odd person out! First, Hollywood has many reasons to be uncomfortable with AI. People from the entertainment industry come from a very different culture than many who work in tech, and this drives deep differences in what we focus on and what we value. A significant subset of Hollywood is concerned that: - AI companies are taking their work to learn from it without consent and compensation. Whereas the software industry is used to open source and the open internet, Hollywood focuses much more on intellectual property, which underlies the core economic engines of the entertainment industry. - Powerful unions like SAG-AFTRA (Screen Actors Guild-American Federation of Television and Radio Artists) are deeply concerned about protecting the jobs of their members. When AI technology (or any other force) threatens the livelihoods of their members — like voice actors — they will fight mightily against potential job losses. - This wave of technological change feels forced on them more than previous waves, where they felt more free to adopt or reject the technology. For example, celebrities felt like it was up to them whether to use social media. In contrast, negative messaging from some AI leaders who present the technology as unstoppable, perhaps even a dangerous force that will wipe out many jobs, has not encouraged enthusiastic adoption. Having said that, Hollywood is under no illusions that AI will change entertainment, and that if Hollywood does not adapt, perhaps some other place will become the new center for entertainment. The entertainment industry is no stranger to technology change. Radio, TV, computer graphics special effects, video streaming, and social media transformed the industry. But the path to navigating AI’s transformation is still unclear, and organizations like the new Creators Coalition on AI are trying to stake out positions. Unfortunately, Hollywood’s negative sentiment toward AI also means it will produce a lot more Terminator-like movies that portray AI as more dangerous than helpful, and this hurts beneficial AI adoption as well. The interests of AI and Hollywood are not always aligned. (Every time I speak in a group like this as the “AI representative,” I can count on being asked very hard questions.) Most of us in tech would prefer a more open internet and more permissive use of creative works. But there is also much common ground, for example in wanting guardrails against deepfakes and a smooth transition for those whose jobs are displaced, perhaps via upskilling. Storytelling is hard. I’m optimistic that AI tools like Veo, Sora, Runway, Kling, Ray, Hailuo, and many others can make video creation easier for millions of people. I hope Hollywood and AI developers will find more opportunities to collaborate, find more common ground, and also steer our projects toward outcomes that are win-win for as many parties as possible. [Original text: https://deeplearning.ai/the-batch/issue-… ]

U.S. policies are driving allies away from using American AI technology. This is leading to interest in sovereign AI — a nation’s ability to access AI technology without relying on foreign powers. This weakens U.S. influence, but might lead to increased competition and support for open source. The U.S. invented the transistor, the internet, and the transformer architecture powering modern AI. It has long been a technology powerhouse. I love America, and am working hard towards its success. But its actions over many years, taken by multiple administrations, have made other nations worry about over reliance on it. In 2022, following Russia’s invasion of Ukraine, U.S. sanctions on banks linked to Russian oligarchs resulted in ordinary consumers’ credit cards being shut off. Shortly before leaving office, Biden implemented “AI diffusion” export controls that limited the ability of many nations — including U.S. allies — to buy AI chips. Under Trump, the “America first” approach has significantly accelerated pushing other nations away. There have been broad and chaotic tariffs imposed on both allies and adversaries. Threats to take over Greenland. An unfriendly attitude toward immigration — an overreaction to the chaos at the southern border during Biden’s administration — including atrocious tactics by ICE (Immigration and Customs Enforcement) that resulted in agents shooting dead Renée Good, Alex Pretti, and others. Global media has widely disseminated videos of ICE terrorizing American cities, and I have highly skilled, law-abiding friends overseas who now hesitate to travel to the U.S., fearing arbitrary detention. Given AI’s strategic importance, nations want to ensure no foreign power can cut off their access. Hence, sovereign AI. Sovereign AI is still a vague, rather than precisely defined, concept. Complete independence is impractical: There are no good substitutes to AI chips designed in the U.S. and manufactured in Taiwan, and a lot of energy equipment and computer hardware are manufactured in China. But there is a clear desire to have alternatives to the frontier models from leading U.S. companies OpenAI, Google, and Anthropic. Partly because of this, open-weight Chinese models like DeepSeek, Qwen, Kimi, and GLM are gaining rapid adoption, especially outside the U.S. When it comes to sovereign AI, fortunately one does not have to build everything. By joining the global open-source community, a nation can secure its own access to AI. The goal isn’t to control everything; rather, it is to make sure no one else can control what you do with it. Indeed, nations use open source software like Linux, Python, and PyTorch. Even though no nation can control this software, no one else can stop anyone from using it as they see fit. This is spurring nations to invest more in open source and open weight models. The UAE (under the leadership of my former grad-school officemate Eric Xing!) just launched K2 Think, an open-source reasoning model. India, France, South Korea, Switzerland, Saudi Arabia, and others are developing domestic foundation models, and many more countries are working to ensure access to compute infrastructure under their control or perhaps under trusted allies’ control. Global fragmentation and erosion of trust among democracies is bad. Nonetheless, a silver lining would be if this results in more competition. U.S. search engines Google and Bing came to dominate web search globally, but Baidu (in China) and Yandex (in Russia) did well locally. If nations support domestic champions — a tall order given the giants’ advantages — perhaps we’ll end up with a larger number of thriving companies, which would slow down consolidation and encourage competition. Further, participating in open source is the most inexpensive way for countries to stay at the cutting edge. Last week, at the World Economic Forum in Davos, many business and government leaders spoke about their growing reluctance to rely on U.S. technology providers and desire for alternatives. Ironically, “America first” policies might end up strengthening the world’s access to AI. [Original text: https://deeplearning.ai/the-batch/issue-…… ]

The full agenda for AI Dev 25 x NYC is ready. Developers from Google, AWS, Vercel, Groq, Mistral AI, SAP, and other exciting companies will share what they've learned building production AI systems. Here's what we'll cover: Agentic Architecture: When orchestration frameworks help versus when they accumulate errors. How model-driven agents and autonomous planning handle edge cases. Context Engineering: Why retrieval fails for complex reasoning tasks. How knowledge graphs connect information that vector search misses. Building memory systems that preserve relationships. Infrastructure: Where hardware, models, and applications create scaling bottlenecks. Semantic caching strategies that cut costs and latency. How inference speed enables better orchestration. Production Readiness: Moving from informal evaluation to systematic agent testing. Translating AI governance into engineering practice. Building under regulatory constraints. Tooling: MCP implementations that work. Context-rich code review systems. Working demos you can adapt for your applications. I'll share my perspective on where AI development is heading. Looking forward to seeing you there!

New course: A2A: The Agent2Agent Protocol, built with @googlecloudtech and @IBMResearch, and taught by Holt Skinner, @ivnardini, and Sandi Besen. Connecting agents built with different frameworks usually requires extensive custom integration. This short course teaches you A2A, the open protocol standardizing how agents discover each other and communicate. Since IBM’s ACP (Agent Communication Protocol) joined forces with A2A, A2A has emerged as the industry standard. In this course, you'll build a healthcare multi-agent system where agents built with different frameworks, such as Google ADK (Agent Development Kit) and LangGraph, collaborate through A2A. You'll wrap each agent as an A2A server, build A2A clients to connect to them, and orchestrate them into sequential and hierarchical workflows. Skills you'll gain: - Expose agents from different frameworks as A2A servers to make them discoverable and interoperable - Chain A2A agents sequentially using ADK, where one agent's output feeds into the next - Connect A2A agents to external data sources using MCP (Model Context Protocol) - Deploy A2A agents using Agent Stack, IBM's open-source infrastructure Join and learn the protocol standardizing agent collaboration! https://deeplearning.ai/short-courses/a2…

Many people are fighting the growth of data centers because they could increase CO2 emissions, electricity prices, and water use. I’m going to stake out an unpopular view: These concerns are overstated, and blocking data center construction will actually hurt the environment more than it helps. Many politicians and local communities in the U.S. and Europe are organizing to prevent data centers from being built. While data centers impose some burden on local communities, most worries of their harm — such as CO2 emissions, driving up consumer electricity prices, and water use — have been inflated beyond reality, perhaps because many people don't trust AI. Let me address the issues of carbon emissions, electricity prices, and water use in turn. Carbon emissions. Humanity’s growing use of computation is increasing carbon emissions. Data-center operations account for around 1% of global emissions, though this is growing rapidly. At the same time, hyperscalers’ data centers are incredibly efficient for the work they do, and concentrating computation in data centers is far better for the environment than the alternative. For example, many enterprise on-prem compute facilities use whatever power is available on the grid, which might include a mix of older, dirtier energy sources. Hyperscalers use far more renewable energy. On the key metric of PUE (total energy used by a facility divided by amount of energy used for compute; lower is better, with 1.0 being ideal), a typical enterprise on-prem facility might achieve 1.5-1.8, whereas leading hyperscalar data centers achieve a PUE of 1.2 or lower. To be fair, if humanity were to use less compute, we would reduce carbon emissions. But If we are going to use more, data centers are the cleanest way to do it; and computation produces dramatically less carbon than alternatives. Google had estimated that a single web search query produces 0.2 grams of CO2 emissions. In contrast, driving from my home to the local library to look up a fact would generate about 400 grams. Google also recently estimated that the median Gemini LLM app query produces a surprisingly low 0.03 grams of CO2 emissions), and uses less energy than watching 9 seconds of television. AI is remarkably efficient per query — its aggregate impact comes from sheer volume. Major cloud companies continue to push efficiency gains, and the trajectory is promising. Electricity prices. Beyond concerns about energy use, data centers have been criticized for increasing electricity demand and therefore driving up electric utility prices for ordinary consumers. The reality is more complicated. One of the best studies I’ve seen, by Lawrence Berkeley National Laboratory, found that “state-level load growth … has tended to reduce average retail electricity prices.” The main reason is data centers share the fixed costs of the grid. If a consumer can split the costs of transmission cables with a large data center, then the consumer ends up paying less. Of course, even if data centers reduce electricity bills on average, that’s cold comfort for consumers in the instances (perhaps due to poor local planning or regulations) where rates do increase. Water use. Finally, many data centers use evaporative cooling to dissipate heat. But this uses less water than you might think. To put this in context, golf courses in the U.S. use around 500 billion gallons annually of water to irrigate their turf. In contrast, U.S. data centers consume far less. A common estimate is 17 billion gallons, or maybe around 10x that if we include water use from energy generation. Golf is a wonderful sport, but I would humbly argue that data centers' societal benefit is greater, thus we should not be more alarmed by data center water usage than golf course usage. Having said that, a shortcoming of these aggregate figures is that in some communities, data center water usage might exceed 10% of local usage, and thus needs to be planned for. Data centers do impose costs on communities, and these costs have to be planned and accounted for. But they are also far less damaging — and more environmentally friendly — than their critics claim. There remains important work to do to make them even more efficient. But the most important point is that data centers are incredibly efficient for the work they do. They have a negative impact because we want them to do a lot of work for us. If we want this work done — and we do — then building more data centers, with proper local planning, is good for both the environment and society. Original text: https://deeplearning.ai/the-batch/issue-…

Job seekers in the U.S. and many other nations face a tough environment. At the same time, fears of AI-caused job loss have — so far — been overblown. However, the demand for AI skills is starting to cause shifts in the job market. I’d like to share what I’m seeing on the ground. First, many tech companies have laid off workers over the past year. While some CEOs cited AI as the reason — that AI is doing the work, so people are no longer needed — the reality is AI just doesn’t work that well yet. Many of the layoffs have been corrections for overhiring during the pandemic or general cost-cutting and reorganization that occasionally happened even before modern AI. Outside of a handful of roles, few layoffs have resulted from jobs being automated by AI. Granted, this may grow in the future. People who are currently in some professions that are highly exposed to AI automation, such as call-center operators, translators, and voice actors, are likely to struggle to find jobs and/or see declining salaries. But widespread job losses have been overhyped. Instead, a common refrain applies: AI won’t replace workers, but workers who use AI will replace workers who don’t. For instance, because AI coding tools make developers much more efficient, developers who know how to use them are increasingly in-demand. (If you want to be one of these people, please take our short courses on Claude Code, Gemini CLI, and Agentic Skills!) So AI is leading to job losses, but in a subtle way. Some businesses are letting go of employees who are not adapting to AI and replacing them with people who are. This trend is already obvious in software development. Further, in many startups’ hiring patterns, I am seeing early signs of this type of personnel replacement in roles that traditionally are considered non-technical. Marketers, recruiters, and analysts who know how to code with AI are more productive than those who don’t, so some businesses are slowly parting ways with employees that aren’t able to adapt. I expect this will accelerate. At the same time, when companies build new teams that are AI native, sometimes the new teams are smaller than the ones they replace. AI makes individuals more effective, and this makes it possible to shrink team sizes. For example, as AI has made building software easier, the bottleneck is shifting to deciding what to build — this is the Product Management (PM) bottleneck. A project that used to be assigned to 8 engineers and 1 PM might now be assigned to 2 engineers and 1 PM, or perhaps even to a single person with a mix of engineering and product skills. The good news for employees is that most businesses have a lot of work to do and not enough people to do it. People with the right AI skills are often given opportunities to step up and do more, and maybe tackle the long backlog of ideas that couldn’t be executed before AI made the work go more quickly. I’m seeing many employees in many businesses step up to build new things that help their business. Opportunities abound! I know these changes are stressful. My heart goes out to every family that has been affected by a layoff, to every job seeker struggling to find the role they want, and to the far larger number of people who are worried about their future job prospects. Fortunately, there’s still time to learn and position yourself well for where the job market is going. When it comes to AI, the vast majority of people, technical or nontechnical, are at the starting line, or they were recently. So this remains a great time to keep learning and keep building, and the opportunities for those who do are numerous! [Original text; https://deeplearning.ai/the-batch/issue-… ]

As amazing as LLMs are, improving their knowledge today involves a more piecemeal process than is widely appreciated. I’ve written before about how AI is amazing... but not that amazing. Well, it is also true that LLMs are general... but not that general. We shouldn’t buy into the inaccurate hype that LLMs are a path to AGI in just a few years, but we also shouldn’t buy into the opposite, also inaccurate hype that they are only demoware. Instead, I find it helpful to have a more precise understanding of the current path to building more intelligent models. First, LLMs are indeed a more general form of intelligence than earlier generations of technology. This is why a single LLM can be applied to a wide range of tasks. The first wave of LLM technology accomplished this by training on the public web, which contains a lot of information about a wide range of topics. This made their knowledge far more general than earlier algorithms that were trained to carry out a single task such as predicting housing prices or playing a single game like chess or Go. However, they’re far less general than human abilities. For instance, after pretraining on the entire content of the public web, an LLM still struggles to adapt to write in certain styles that many editors would be able to, or use simple websites reliably. After leveraging pretty much all the open information on the web, progress got harder. Today, if a frontier lab wants an LLM to do well on a specific task — such as code using a specific programming language, or say sensible things about a specific niche in, say, healthcare or finance — researchers might go through a laborious process of finding or generating lots of data for that domain and then preparing that data (cleaning low-quality text, deduplicating, paraphrasing, etc.) to create data to give an LLM that knowledge. Or, to get a model to perform certain tasks, such as use a web browser, developers might go through an even more laborious process of creating many RL gyms (simulated environments) to let an algorithm repeatedly practice a narrow set of tasks. A typical human, despite having seen vastly less text or practiced far less in computer-use training environments than today's frontier models, nonetheless can generalize to a far wider range of tasks than a frontier model. Humans might do this by taking advantage of continuous learning from feedback, or by having superior representations of non-text input (the way LLMs tokenize images still seems like a hack to me), and many other mechanisms that we do not yet understand. Advancing frontier models today requires making a lot of manual decisions and taking a data-centric AI approach to engineering the data we use to train our models. Future breakthroughs might allow us to advance LLMs in a less piecemeal fashion than I describe here. But even if they don’t, the ongoing piecemeal improvements, coupled with the limited degree to which these models do generalize and exhibit “emergent behaviors,” will continue to drive rapid progress. Either way, we should plan for many more years of hard work. A long, hard — and fun! — slog remains ahead to build more intelligent models. [Original text: https://deeplearning.ai/the-batch/issue-…… ]