Bin Yu, PhD’s Post

Highly recommend “Understanding Deep Learning," authored by Simon Prince and published in 2023. This comprehensive guide is perfect for both practitioners and students. To further enrich your learning experience, the book's website offers 68 Python notebook exercises, providing practical programming practice. Dive into deep learning today and elevate your skills with "Understanding Deep Learning." Explore more at udlbook.github.io/udlbook/.

Here's a compelling article from Sequoia Capital, arguing that the tech industry needs to generate $600 billion in AI revenue to justify the significant investments in GPUs and data centers. Currently, in the ToC market, OpenAI stands as the largest dedicated AI company, with an annual run rate of $3.4 billion. However, this situation appears unsustainable unless a valuable product emerges that a lot of consumers are willing to purchase, like those services from Netflix or Spotify. https://lnkd.in/e8cX475a

I’m happy to share that I’m starting a new position as Founder and CEO at GrowthOps Solutions LLC! GrowthOps specializes in consulting services for achieving profitable and sustainable business growth through data and AI (including Gen AI) strategies and solutions.

The article discusses the challenges firms face in transitioning generative AI proofs of concept (PoCs) into profitable ventures, highlighting issues such as high implementation costs, talent acquisition, data availability, accuracy and privacy concerns. Despite significant investments, many companies struggle to see substantial returns from their AI initiatives. For many companies, it is an efficiency play now. There are not many use cases that can demonstrate revenue generation compared to traditional ML models, such as search and recommendation. Without a well defined ROI, the investment to generative AI at many companies is not sustainable.

GraphRAG (Graph-based Retrieval Augmented Generation) improves upon traditional Retrieval Augmented Generation (RAG) by incorporating knowledge graphs (KGs) or graph databases with large language models (LLMs). This method utilizes the structured nature of graph databases, which organize data into nodes and relationships, to facilitate more efficient and accurate retrieval of relevant information. By providing better context to LLMs, GraphRAG enhances response generation. Integrating knowledge graphs as sources of structured, domain-specific context or factual information allows GraphRAG to deliver more precise, contextually aware, and relevant answers. This is particularly effective for complex queries requiring a comprehensive understanding of summarized semantic concepts across large data collections or single extensive documents. https://lnkd.in/eJfXRiah

Yvonne Yeung is hiring. It would be a great opportunity for you to join a top notch team and contribute to the Starbucks digital ecosystem for stores and customers.

OpenAI, Stanford, Microsoft, Princeton University and other institutions selected 1,565 papers from numerous research reports Released the large-scale model "Prompt Technical Report" The research team screened more than 1,500 relevant papers from various academic resources, compiled a set of standard terms and prompts, and created a directory that is easy to understand and use. https://lnkd.in/e-QqwjSy

The cost of training large Gen AI models has surged from millions to tens of millions of dollars. This rapid increase is due to the escalating complexity and scale of these models, requiring more advanced hardware and greater computational power. For example, training GPT-4 last year cost around $78.4 million, while Google's Germini Ultra reached a staggering $191 million. This graph illustrates the steep rise in training costs over time.

The picture from Gartner indicates that generative AI is highly effective in content generation (including text, image, and video creation, as well as synthetic data) and conversational user interfaces (such as virtual assistants, chatbots, and digital workers). It shows moderate effectiveness in segmentation/classification tasks (like clustering and customer segmentation) and recommendation systems (providing personalized advice and next best actions). However, generative AI currently has low effectiveness in prediction/forecasting (including risk prediction and sales forecasting) and decision intelligence (such as decision support and automation). This delineation highlights where generative AI is most and least useful.