Smart Talks With IBM: Transformations in AI: why foundation models are the future

    enSeptember 20, 2023

    Podcast Summary

    • Exploring the Impact of AI on Businesses and SocietyIBM's VP of AI models, David Cox, discusses the transformative potential of AI for businesses, from innovative workflows to groundbreaking discoveries, and shares insights into foundation models and generative AI during an interview on the Smart Talks with IBM podcast.

      The Kakadu plum, an Australian native superfood, contains 100 times more vitamin C than oranges, yet it's not widely known. Meanwhile, the tech world is abuzz with the transformative potential of artificial intelligence (AI), a topic explored in the new season of the Smart Talks with IBM podcast. AI, explained IBM's VP of AI models, David Cox, is a game-changer for businesses, with practical applications ranging from innovative workflows to groundbreaking discoveries. The IBM-MIT Watson AI Lab, a five-year-old industry-academic collaboration, is at the forefront of AI research. Cox, who has dedicated his life to AI, shared insights into foundation models and generative AI during his interview with Jacob Goldstein. The partnership between IBM and MIT, he noted, has a rich history in AI research. While the world may not yet be sipping Kakadu plum smoothies, the potential impact of AI on businesses and society is undeniable. Tune in to the Smart Talks with IBM podcast for more insights on this revolutionary technology.

    • IBM-MIT partnership driving AI innovation since 1956IBM's collaboration with MIT has been instrumental in driving AI research and innovation for over 65 years, enabling IBM to stay ahead of the curve and explore broader applications of AI.

      The partnership between IBM and MIT, which began at the inception of artificial intelligence in 1956, has been instrumental in driving IBM's AI strategy. With IBM committing over $1 billion to this joint lab located on MIT's campus, both entities have been able to collaborate closely, allowing them to work together on the latest AI research and projects. This unique relationship enables IBM to stay ahead of the curve in areas like foundation models and explore broader applications of AI, such as addressing climate change and creating advanced materials. Recently, there's been a surge in public interest in AI due to breakthroughs in generative AI models like ChatGPT and Midjourney. Some of the key advancements in AI that have fueled this excitement include the algorithm called backpropagation, which has been in use since the 1980s but gained momentum with the availability of large amounts of data and compute power. Overall, the IBM-MIT collaboration has been a driving force in the field of AI, allowing for the exploration of new ideas and applications that benefit both society and IBM's customers.

    • Recent AI advancements fueled by data, powerful computers, and self-supervised learningThe availability of massive amounts of data, powerful computers, and self-supervised learning techniques have led to recent breakthroughs in AI, including the development of foundation models that offer a base for further development and the ability to use unannotated data for training.

      The recent advancements in artificial intelligence (AI), particularly in deep learning and self-supervised learning, are largely due to the availability of massive amounts of data, powerful computers, and the ability to use unannotated data. These technologies, which have been simmering for some time, were brought to the public's attention with the release of models like ChatGPT and those using self-generative technologies like stable diffusion and mid-journey. These models, often referred to as foundation models, provide a foundation or base for further development by allowing users to build upon their capabilities with minimal additional training. The ability to use unannotated data in self-supervised learning has been a game-changer, allowing for the use of vast amounts of data and leading to more powerful models. The public's awareness of these technologies has been significantly increased by their ability to interact with them, leading to a surge in interest and excitement. Foundation models, which offer a foundation for further development, have been a key factor in this recent surge in AI advancements.

    • From specialized AI appliances to versatile foundation modelsFoundation models reduce the effort and cost of automation, enabling access to value from a wider range of data sources and increasing efficiency.

      Foundation models represent a significant shift in the field of AI, moving from single-use appliances to a versatile oven with a range top. Before foundation models, the effort required to automate a task was extremely high due to the need for large amounts of carefully labeled data and skilled labor. This high cost made it necessary to only invest in automating high-value tasks. Now, with foundation models, the effort required to achieve results has been dramatically reduced, allowing for a wider range of applications and more efficient use of resources. Think of traditional AI solutions as building a dam on a river of data. Dams are expensive and require specialized skills, making it necessary to only build them on large rivers that will provide a significant return on investment. However, the vast majority of the water in a kingdom is not in the river but in puddles, creeks, and babbling brooks. Foundation models allow us to access and utilize this previously untapped water, providing value from a wider range of data sources and reducing the overall cost of automation. In essence, foundation models have lowered the barrier to entry for automation, making it possible to extract value from a greater variety of data sources and reducing the need for extensive upfront investment and labor. This shift has the potential to unlock significant value and drive innovation across a wide range of industries and applications.

    • Choosing the right-sized foundation modelFoundation models offer a strong base for automating tasks, but it's crucial to consider their size and specialization for optimal performance and cost-effectiveness.

      Foundation models offer a significant advantage in automating new tasks by providing a strong base of knowledge, reducing the effort required compared to starting from scratch. However, it's essential to recognize that there isn't a one-size-fits-all foundation model. While these models can be powerful and relatively easy to build upon, they consume a lot of energy and come in various sizes. Choosing the right-sized model for specific problems can lead to better performance and efficiency. For instance, a smaller model specialized in a particular domain may outperform larger models. Stanford's example of a 2.7 billion parameter model trained on biomedical literature demonstrates this, as it was more effective and efficient than larger models. In summary, foundation models provide a solid foundation for automating tasks, but it's crucial to consider the size and specialization of the model for optimal performance and cost-effectiveness.

    • Addressing bias and hallucination in AIResearchers are tackling bias and hallucination in AI, focusing on methods to remove biases, identify them, and create tools for auditing systems to ensure fairness and accuracy.

      As the use of AI continues to expand, there will be a tension between vendors promoting large, all-encompassing models and those advocating for a more nuanced approach, selecting tools that best fit specific tasks. A significant challenge in AI is addressing bias and hallucination. Bias, a long-standing issue, arises when models are trained on biased data, resulting in biased outputs. For instance, a translation system might translate "they are a nurse" to "she is a nurse," reflecting societal biases. Researchers are actively working on methods to remove biases, identify them, and create tools for auditing systems. Another concern is hallucinations, where models generate false information. For example, Chatty BT, a language model, created a fake biography for the speaker, claiming they were British despite being American. These issues, particularly relevant to foundation models and large language models, require ongoing attention and solutions to ensure fairness and accuracy.

    • Challenges and Opportunities of AI for BusinessesAI offers potential for business applications like text analysis, summarization, and question answering, but businesses must address challenges posed by AI's ability to generate fictional information.

      While chatbots and generative AI like ChatGPT can be entertaining and useful for personal use, they present challenges for businesses due to their ability to invent fictional information with confidence. This can lead to misunderstandings or incorrect data being used, creating gaps that need to be addressed. However, these technologies also offer significant potential for business applications, particularly in areas such as text analysis, summarization, and question answering. These capabilities can streamline processes, save time, and improve efficiency in various business contexts. For instance, sentiment analysis of product reviews, summarization of long texts like customer complaints, and quick question answering can all be valuable for businesses. Overall, the challenge lies in harnessing the strengths of AI while mitigating its weaknesses to effectively address business needs.

    • The future of human-computer interaction: conversational interfacesConversational interfaces allow users to interact with systems using everyday language, revolutionizing productivity and impacting industries like HR and coding.

      The future of human-computer interaction is shifting towards more natural, conversational interfaces. This technology, which allows users to interact with systems using everyday language, is set to revolutionize productivity by making it easier for us to ask questions, give commands, and even generate code. This conversational interface is a departure from the traditional graphical user interface (GUI), which has been the standard for decades. In fact, conversing with machines could become as commonplace as using a GUI. This technology has the potential to greatly impact various industries, including HR, where employees can now ask about their vacation days using natural language. Furthermore, in the realm of coding, conversational interfaces could lead to a significant revolution, enabling users to describe what they want, and AI generating the code for them. The possibilities for AI applications in business are vast, with some unique use cases including the development of linkage systems in mechanical engineering, where AI can help design and build complex machines by understanding and interpreting human language.

    • AI is transforming industries and creating new opportunitiesAI is making processes more efficient and productive, leading to cost savings, new opportunities, and customer delight. While some jobs may be displaced, history shows that new opportunities will emerge as technology transforms the way we work.

      AI technology is revolutionizing various industries by automating complex tasks that were previously beyond human capability. From building intricate machinery to generating electronic circuits, AI is making processes more efficient and productive. For businesses, this means cost savings, new opportunities, and customer delight. For employees, while some jobs may be displaced, history shows that new opportunities will emerge as technology transforms the way we work. Just as automation led to the decline of agriculture jobs and the rise of knowledge work, AI is expected to continue making us more productive, leading to an insatiable appetite for doing even more. While concerns about job displacement are valid, history suggests that people will transition to new roles as technology advances.

    • IBM's Watson X: Making AI Accessible for BusinessesIBM's Watson X simplifies AI adoption for businesses through accessible tools and foundation models, enabling innovation and competitiveness in the rapidly evolving AI landscape.

      IBM's new offering, Watson X, represents a shift from labor-intensive automation to a world of easier AI usage for businesses. Watson X is a new branding on IBM's Watson AI technologies, bringing together tools for businesses to harness the power of AI more effectively. This includes foundation models and data components. Creativity plays a significant role in this field, from inventing new solutions to identify and solve real-world problems, to matchmaking between technology and solvable issues. Watson X aims to help businesses manage the risks and complexities of enterprise AI adoption, making it easier for them to innovate and stay competitive in the rapidly evolving AI landscape. The creative process in AI research involves identifying problems, building new solutions, and understanding the ever-changing possibilities of what AI can achieve. The environment for this work is exciting and collaborative, with researchers drawing on whiteboards, writing on windows, and constantly pushing the boundaries of what's possible.

    • Imagining AI as a natural resource for transformationIn 20 years, AI could be a seamless part of our lives, offering opportunities for innovation, productivity, and enrichment, rather than a job displacer. Understanding its transformative potential is crucial.

      AI is imagined to work in harmony with people 20 years from now, much like a natural resource. This means we could be surrounded by seamless automation and intellectual capabilities, leading to an abundance of opportunities for innovation, productivity, and enriching our lives. AI isn't necessarily a job displacer, but rather a powerful resource to augment our abilities. This vision, as described by MIT economist David Autor, offers a transformative potential for our future. Smart Talks at IBM, featuring David Cox, provides a deep exploration of this topic, emphasizing the importance of understanding AI's transformative potential and its role in our lives. This conversation highlights the potential for natural conversation between mankind and machine to generate creative solutions to complex problems, ultimately boosting our innovation and productivity. We eagerly await the next breakthroughs in AI, as described by David Cox, and the endless possibilities it brings.

    Recent Episodes from Stuff To Blow Your Mind

    From the Vault: Do ants make traps?

    From the Vault: Do ants make traps?

    The trap is insidious. But of course it is. It was designed and executed… by ants? In this episode of Stuff to Blow Your Mind, the ant wars continue as Robert and Joe explore some of the possibility that a few formidable Formicidae species actually lay traps. (originally published 01/20/2022)

    See omnystudio.com/listener for privacy information.

    Related Episodes

    EP 243: 5 Simple Ways To Use Generative AI Every Day

    EP 243: 5 Simple Ways To Use Generative AI Every Day

    If I had to count, I probably use Generative AI at least 100 times a day.  Some are advanced. Some are super simple. I'm going to be sharing some of the simplest ways that I use Generative AI to create Everyday AI and tell you the ways that you can use my same tactics to start winning back time today.
    Newsletter: Sign up for our free daily newsletter
    More on this Episode: Episode Page
    Learn more in today's newsletter
    Join the discussion: Ask Jordan questions on AI

    Related Episodes:
    Ep 189: The One Biggest ROI of GenAI
    Ep 197: 5 Simple Steps to Start Using GenAI at Your Business Today

    Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineup
    Website: YourEverydayAI.com
    Email The Show: info@youreverydayai.com
    Connect with Jordan on LinkedIn

    01:30 Daily AI news
    07:17 Knowledge workers adapt to succeed in AI era.
    10:44 Behind the scenes of Everyday AI production.
    18:21 Steal this method to learn about your career.
    19:10 Customize Google search for efficient news updates.
    23:01 Using shortcuts to gather information efficiently.
    27:15 Using CastMagic and its features.
    29:13 Automated transcription
    33:57 Microsoft 365 Copilot helps with team work.
    37:05 Addicted to Chrome, but now liking Edge.
    40:20 Request to find specific generative AI statistics in PDF.
    44:17 Chat prompt reused for specific AI training.
    45:02 Optimize chat model use with refined method.
    50:12 Use perplexity for efficient research queries, not Google.
    52:03 Generative AI can save knowledge workers time.

    Topics Covered in This Episode:
    1. Practical uses of Generative AI
    2. Tips for saving time using AI
    3. Summarizing web pages for quick updates
    4. Organizing and training ChatGPT for specific tasks

    Microsoft 365 Copilot, Microsoft 365 Teams, Cast Magic, Chat GPT, Microsoft Edge, Chromium, Chrome Extensions, Copilot Integration, Perplexity, Google Search, Generative AI, McKinsey Digital, AI News, Amazon Web Services, Voila, Organizing Chat GPT, AI in Research, Custom Prompts, Knowledge Workers, New York City AI-powered Chatbot, Google's AI Reply Suggestions, Premium AI Features, AI Automation, Chat GPT Summaries, AI Content Generation, Everyday AI, AI News Headlines, Third-Party Information Connection, Bullet Points Summaries, Web Content Summarization.

    Get more out of ChatGPT by learning our PPP method in this live, interactive and free training! Sign up now: https://youreverydayai.com/ppp-registration/

    Navigating the Chatbot Jungle: Cyber Mindful Conversations

    Navigating the Chatbot Jungle: Cyber Mindful Conversations
    Unveil the fascinating world of AI chatbots with Sandra Estok on this enlightening episode of Cyber Mindful Podcast. From the promises of time-saving interactions to the potential pitfalls of data exploitation, Sandra navigates the intricate landscape of artificial intelligence. Real-life stories underscore the importance of responsible usage and safeguarding personal data.

    Join the conversation as Sandra demystifies AI chatbots, likening them to interactive companions within our devices. Explore the benefits and frustrations of these digital assistants, gaining insights into their vast potential and occasional limitations. The narrative takes a cautionary turn with stories of data exploitation, emphasizing the need for mindful and intentional engagement in the digital realm.

    In this episode, Sandra equips listeners with practical tips for secure chatbot interactions. From verifying platform legitimacy to avoiding oversharing, these rock-solid guidelines ensure a safer and more intentional digital experience.

    Sandra Estok is the founder of Way2Protect™, author of the international bestselling “Happily Ever Cyber!™” book, and the Cyber Literacy series for children. With over 25 years of experience in Cybersecurity, IT, and Data Privacy, Sandra is a Keynote Speaker and Corporate Trainer. Her mission is to empower women to take charge of their cyber safety and have Peace of Mind Online.

    Connect with Sandra:
    Website SandraEstok.com
    Social media handle: @Way2Protect

    Watch Sandra Estok's TEDx talk “What an identity theft victim can teach about cybercrime”: here. Or copy this link directly into your browser: https://www.ted.com/talks/sandra_estok_what_an_identity_theft_victim_can_teach_us_about_cybercrime

    The Future of Insurance – Juan de Castro, COO, Cytora

    The Future of Insurance – Juan de Castro, COO, Cytora

    Juan de Castro is COO of Cytora, leading product strategy, sales, distribution, partnerships and operations. He was previously UK COO at Hiscox (FTSE100), and lead Group Strategy and Corporate Development at the Group level, reporting to the Group CEO.

    Before moving into Insurance, Juan was a senior manager at McKinsey’s Silicon Valley office, focusing on growth strategy and operational improvement engagements for Fortune 100 companies. He worked across a number of different industries, from Tech to Utilities and Financial Services, with a special focus on digital strategy and operational turnarounds. 

    A Computer Engineer by training, he started with a technical career owning P&L responsibility in a professional services division, building and leading engineering teams. He holds a Master’s degree from Stanford Graduate School of Business and a Computer Engineering degree from Universidad Complutense de Madrid.

    Highlights from the Show

    • Juan came to Insurance from consulting in Silicon Valley, and admits he didn't realize how much he didn't realize that justified the sense some have of the industry being behind, like not knowing the cost of the goods you're selling when you sell it
    • He also found that we spend a lot to bring customers in, only to decide they're out of appetite, especially in the direct channel, where his carrier work focused
    • Serving as COO of Hiscox UK, he focused a lot on the expense ratio, but worked on driving up premium per underwriter rather than cutting costs, which he sees as less sustainable or impactful
      • He found getting the data you really need to do this is hard, like where underwriters spend their time or lose it
      • Once you had the data, you had limited tools to deliver on what you discovered, like how to identify the right risks quickly and then get them to the right underwriters
    • Juan sees a growing focus on the end-to-end workflow of underwriters, much like how manufacturing transformed in the Industrial Revolution
      • New tools enabled the flow of work and automation
      • New energy enabled speed and deploying tools that weren't possible before
      • Automation allowed the new fuel and tools to push product through the system
      • This is actually very similar to how underwriters work, and we're in the midst of the same type of drivers coming online
        • New tools, like AI and Gen AI
        • New fuel, meaning better internal and external data
        • New ability to deploy automation that didn't exist before, so risk can flow (which is the idea behind Cytora's podcast that Juan hosts, called Making Risk Flow)
      • This is a moment where underwriters can rise above the transaction to focus on macro-level issues and engage with brokers
    • In the mid-market area that Juan and Cytora focuses, he talks about 10 Click Underwriting
      • This is about using tools to remove the excessive number of steps and handoffs internally so risks can come in and be quoted within hours instead of days or weeks
      • This includes using third party data and artificial intelligence to take in submissions, analyze and prioritize them in conjunction with a rating engine, to present prioritized opportunities to underwriters with the transactional work done for them
    • When it comes to AI, Juan is interesting in enabling both the early adopters and those who are on the sidelines, unsure of what to do
      • You need to have the right controls and safeguards, like not sending proprietary data to the LLMs to protect carrier's IP
      • LLMs allow carriers to overcome historic roadblocks to digitization, with three examples he shared:
        • Ingesting submissions from brokers and taking the insights from them without having to train models on hundreds or thousands of examples
        • Mapping data, once extracted, so your internal systems can understand, like taking in varied occupancy reports in a way that informs a rating engine and record of risk system without a person having to make sense of it
        • Bring together UW guidelines and the book of business to help guide underwriters on how to handle a specific risk – something you'd need years of coaching, shadowing and training to achieve otherwise
      • Juan thinks the horizon for adoption of AI is different from past technology like IoT because the benefits aren't theoretical, they're real, tangible and demonstrable, so he believes the industry will move faster here

    This episode is brought to you by The Future of Insurance book series (future-of-insurance.com) from Bryan Falchuk.

    Follow the podcast at future-of-insurance.com/podcast for more details and other episodes.

    Music courtesy of Hyperbeat Music, available to stream or download on Spotify, Apple Music, and Amazon Music and more.

    Ep84- The Great Equalizer: Gen AI and AI Transforming the Data Center

    Ep84- The Great Equalizer: Gen AI and AI Transforming the Data Center
    The generative AI wave is lifting all boats, including those in the data center. In this podcast, we learn how generative AI and automation lower the bar for data center management. In addition to discussing why Gen AI is so great for the data center, we also explore why the data center is also so valuable for Gen AI and why IT should be driving AI efforts.