DeepLearning.AI

AI Dev 25 is officially sold out... almost! Regular tickets are gone, but we’ve opened up a limited batch of last chance and afternoon passes for those who still want to join us on Pi Day (3.14.25) in San Francisco! 🎟️ Grab your spot now before these sell out too: https://bit.ly/4aNM7he

Researchers at Carnegie Mellon University introduced a tree search method for language model agents, improving their ability to complete tasks using the web. By treating interactions as a branching search, agents can evaluate multiple action paths, avoid repeating mistakes, and backtrack when necessary. In tests where the agent navigated simulated online shopping, forums, and classified ads, agents using tree search successfully completed 26.4% of tasks, outperforming a baseline that lacked search capabilities. Read our summary of the paper in The Batch: https://hubs.la/Q036Q-jh0

And then realizing you forgot to actually call the function. ¯\_(ツ)_/¯ . . . . . Original seen in Programmer Humor on Reddit.

Google revised its AI principles, removing previous commitments to avoid work on weapons, surveillance, and other military applications beyond non-lethal uses. The company's updated position, detailed in its 2025 Responsible AI Progress Report, aligns with a broader industry shift as OpenAI, Meta, and Anthropic also eased restrictions on defense applications. In its new AI Principles document, Google emphasizes AI’s role in national security, scientific progress, and economic growth, stressing responsible development and the importance of democratic values. Learn more in The Batch: https://hubs.ly/Q036QTjl0

How do we measure the accuracy of AI models in medicine? In diagnostic AI, sensitivity and specificity are key—sensitivity tells us how well a model detects disease, while specificity measures how well it avoids false positives. These metrics are critical for evaluating real-world AI applications in healthcare. Learn how to build and assess AI models for medical diagnosis in the AI for Medicine Specialization: https://hubs.la/Q036RhRm0

A strong foundation is key to a successful AI career. In this classic letter, Andrew Ng stresses the importance of building solid ML knowledge—and how it underpins everything from housing-price predictors to the most advanced deep learning. Read his letter here: https://hubs.la/Q036phhl0

Shapley values provide a way to fairly distribute credit (or blame) among features, showing their impact on a model’s output. Whether predicting patient risk or explaining AI decisions, these values help unpack how models make decisions. Explore this concept and more in the AI for Medicine Specialization: https://hubs.la/Q036R7qX0

Between a full-time job and a passion for AI, Alok Singh took on the AI for Medicine Specialization—learning how to detect diseases from X-rays, predict heart attack risk, and segment MRI images. In his words: 'This course turned out to be the one that truly hooked me.' Read about his experience, insights, and why he recommends this specialization for anyone serious about AI in healthcare: https://hubs.la/Q036Q-jq0 Get started yourself here: https://hubs.la/Q036Rd_r0

Alibaba released Qwen2.5-VL, a family of open vision-language models available in three sizes (3B, 7B, and 72B parameters). The models process images, text, and videos and can also interact with computer interfaces. Qwen2.5-VL-72B outperformed competitors like GPT-4o and Gemini 2.0 Flash on key benchmarks. The Chinese giant also launched Qwen2.5-Max, a mixture-of-experts model, and Qwen2.5-1M, a family of smaller language models with a 1M-token context window. Learn more in The Batch: https://hubs.la/Q036QVn70

Google Cloud is bringing multimodal AI learning to AI Dev 25! 📝🖼️🎥 On March 14, join Paige Bailey for a hands-on workshop: A Beginner's Guide to Multimodal AI with Gemini 2.0, Veo 2, and Imagen 3 in AI Studio. Get an inside look at how these models can generate text, images, and videos—all in one place with AI Studio. More workshops to be announced soon! You can still get afternoon-only tickets here: https://lnkd.in/e5-GXM5s See you on Pi Day in San Francisco 🌁