x-andrewyng

Andrew Ng highlights the importance of "permissionless innovation," a concept from Adam Thierer's book, arguing that the best ideas emerge without requiring government pre-approval. He states that protecting open source AI is now crucial to preserving this principle, especially as debates over AI governance and potential licensing regimes intensify.

Andrew Ng describes three loops for AI-assisted product development: the agentic coding loop (AI iterates on code), the developer feedback loop (human steers product), and the external feedback loop (user testing informs vision). He notes engineers are taking on product management roles as AI accelerates development.

Anthropic restricted use of its Claude Fable 5 model for building competing LLMs, and the U.S. imposed export controls, leading Anthropic to disable global access. Andrew Ng warns this accelerates efforts by nations and businesses to secure independent AI access and invest in open-source alternatives.

Andrew Ng announced a new DeepLearning.AI course on adding voice to AI agents, built with VocalBridge and taught by its CEO. The course teaches how to build fast and reliable voice applications, covering voice-interactive games, adding voice to agents without rewriting prompts, and outbound phone call agents. It also covers voice evaluation for quality assurance before deployment.

AI Forward Deployed Engineers (FDEs) are a growing role in Silicon Valley, with OpenAI and Anthropic building teams to embed engineers within client organizations to customize AI solutions. Andrew Ng argues the demand for generalist AI Engineers will far exceed that for FDEs, and predicts AI engineering will eventually fragment into specialized roles like LLMOps and Evals Engineers.

Andrew Ng announced a new short course built with Red Hat on serving LLMs efficiently. The course covers quantization to reduce memory footprint, serving models with vLLM for concurrent requests, and benchmarking to balance speed, cost, and accuracy.

Andrew Ng shares a Memorial Day message honoring those who sacrificed everything to defend democracy, noting that democracy is not guaranteed but requires constant care as a precious inheritance.

The new White House policy requires green card applicants to apply from outside the US, which the author argues is a capricious attack on legal immigration that will harm families, reduce the number of doctors, teachers, and scientists, and hurt American competitiveness in AI.

Andrew Ng criticizes Harvard's decision to cap A grades at 20%, arguing education should help all students succeed. He says GPA is not a vital hiring signal and prefers defining elite education by enabling broad excellence rather than limiting success.

Andrew Ng's DeepLearning.AI launched a course on AI agents for image and video generation, built with Google Cloud. It teaches three evaluation techniques—image-text similarity, LLM judging, and structured rubrics—for agents to self-improve output quality.

Andrew Ng announced a new "Transformers in Practice" course built in partnership with AMD. The course offers a practical view of transformer-based LLMs, covering token generation, attention mechanisms, inference bottlenecks, and quantization—using interactive visualizations to build intuition.

Andrew Ng argues the AI jobpocalypse narrative is overblown, noting strong software hiring and low unemployment. He says AI will create more jobs than it destroys, predicting an "AI jobapalooza" instead.

Coursera and Udemy have merged into one company, combining their educational content and platforms to serve learners globally. Andrew Ng, founder of both companies, will serve as Chairman of the combined entity, which aims to address the growing need for continuous learning driven by AI and changes in the nature of work.

Andrew Ng announced a new short course, built with CopilotKit, on building agents that generate custom UIs (charts, forms, whiteboards) in real-time within chat. The course covers three approaches: using pre-built components, composing layouts from building blocks, and integrating third-party apps directly into conversations.

Andrew Ng ranks coding agent acceleration from most to least effective: frontend development (dramatically sped up), backend (slower due to bugs and security), infrastructure (limited due to complex tradeoffs), and research (marginal help beyond coding). He uses this categorization to adjust team expectations and organization.

Andrew Ng explains that AI-native software engineering teams operate differently from traditional teams, using coding agents to build products much faster. This speed shifts bottlenecks to product management, marketing, and legal compliance, forcing engineers and PMs to broaden their roles. He argues that small, co-located teams with generalist skills can move fastest in this new paradigm.

Andrew Ng writes from the World Economic Forum that businesses need to move beyond incremental AI efficiency gains to achieve transformative impact. He argues that workflow redesign, rather than isolated AI projects, is key to creating significant business value, using loan processing as an example where end-to-end changes can transform product offerings.

A new course called "Agent Skills with Anthropic" teaches how to create custom skills for AI agents using Anthropic's open standard format. The course covers building skills for code generation, data analysis, and research, and deploying them across Claude platforms.

U.S. policies are driving allies to seek sovereign AI capabilities to reduce reliance on American technology. This has led to increased interest in open-source AI models and domestic foundation models from countries like the UAE, India, and France. While weakening U.S. influence, this trend may spur greater competition and support for open-source alternatives.

Job seekers face a tough environment, but widespread AI-caused job losses have been overblown so far. Instead, AI is creating demand for workers with AI skills, while those who don't adapt may face replacement. Companies are building smaller, AI-native teams as AI makes individuals more productive.

A new course teaches A2A, an open protocol standardizing how AI agents discover and communicate with each other. The course covers building healthcare multi-agent systems where agents from different frameworks collaborate through A2A. Students learn to expose agents as A2A servers, chain them sequentially, connect to external data sources, and deploy using IBM's Agent Stack.

Andrew Ng spoke at the Sundance Film Festival about AI's impact on Hollywood. He noted Hollywood's concerns about AI using creative work without consent and threatening jobs, but acknowledged the industry must adapt to technological change. Ng expressed hope for collaboration between Hollywood and AI developers to find common ground.

Andrew Ng sends a Valentine's Day message to his AI friends, saying they raise his temperature parameter. The tweet includes a heart emoji for the holiday.

Andrew Ng discusses how AI can create new job opportunities, using the example of designing a birthday cake with AI that required a baker to create it. He argues that while AI may displace some jobs, it also generates demand for new roles and services, potentially increasing overall employment in creative and technology sectors.

Inception Labs has launched Mercury 2, described as the world's first reasoning diffusion LLM. The diffusion language model reportedly delivers 5x faster inference speed compared to leading speed-optimized LLMs.

A new course teaches how to build and train a 20-million parameter language model from scratch using JAX, the open-source library behind Google's Gemini and Veo models. The course covers implementing a MiniGPT-style architecture, training it, and interacting with the finished model through a chat interface.

Apple named its latest laptop Neo, which shares the name of the author's son. The author questions whether to purchase one and jokes about running Amazon Nova on it to impress both children.

Andrew Ng announces Context Hub, an open tool that provides coding agents with up-to-date API documentation via a CLI. The tool aims to prevent agents from using outdated APIs and hallucinating parameters, and is designed to improve over time as agents annotate documentation and share learnings.

Andrew Ng announced Context Hub (chub), an open CLI tool that provides coding agents with up-to-date API documentation. The tool has gained over 6,000 GitHub stars and scaled to over 1,000 API documents through community contributions. The new release allows agents to share feedback on documentation to help refine it for everyone.

DeepLearning.AI has launched a new short course called "Agent Memory: Building Memory-Aware Agents" in partnership with Oracle. The course teaches how to build memory systems that enable AI agents to persist and retrieve information across sessions, allowing them to learn over time. It covers designing memory managers, implementing semantic tool retrieval, and building write-back pipelines.