[DSC DACH 23] ChatGPT and Beyond: How generative AI is Changing the way people interact with computers - Iveta Lohovska

Editor's Notes

• #3: Intent of Slide: Explain why exascale era technologies are needed in digital transformation Massive data growth felt by every organization, big and small is driving the need for larger and more detailed models and simulations and new ways to compute on that data with new algorithms such as AI and other data models that are used together with new algorithms that push the limits of loosely-coupled, scale out systems. Taken together, this requirements are driving the need for infrastructure that increasingly relies on HPC and supercomputing technologies. We feel in time, that as digital transformation continues to take hold—all datacenters will require supercomputing technologies to support these new modeling, simulation, analytics, and AI workflows.
• #4: Intent of the slide: Illustrate HPE engagement with Zenseact for Highly Autonomous Driving If there’s an auto brand synonymous with driver safety, it’s Volvo. And if there’s a new technology pushing the automobile into the vast unknown, it’s autonomous driving. At the center of that crossroads is Zenseact—a Volvo Cars-owned start-up that uses an HPE built consumption-based platform to deliver thousands of simulations per second and make cars safer. The challenge of autonomous driving is one of the greatest facing the automotive industry today. To develop software that will safely and securely allow cars to drive autonomously, huge amounts of data needs to be collected and processed. Zenseact (formerly Zenuity), which is today wholly owned by the Swedish car maker, Volvo Cars Corporation (VCC) , has designed their platforms to deliver world-class performance in autonomous driving, while at all times meeting stringent real-world safety benchmarks. HPE will provide the entire HPC/AI infrastructure and services to support the development of next generation autonomous driving features (L3), available on the market from 2023. Press Release https://www.hpe.com/psnow/doc/a50004350enw For more information, please, contact [email protected]
• #8: Pause Giant AI Experiments: An Open Letter We call on all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4.
• #14: High-resolution image reconstruction with latent diffusion models from human brain activity People had their brains scanned in an fMRI machine while viewing a target image. The scan data was then decoded and reconstructed with a diffusion model.
• #15: Intent of the slide: Expand Trustworthy AI (relative to all AI conversations) Trustworthy AI is AI where accuracy, equity, sustainability, privacy, compliance, and confident are engineered in. As more companies adopt AI, the importance of embedding key principles that enforces ethical and inclusive decisions in AI designs and models is critical. HPE’s ethical guidance for AI serves two critical needs: Demonstrates with transparency how current technology can be applied with confidence Illuminates where current technologies fall short and that’s where we need to innovate. At HPE, we’ve crafted 5 principles to guide ethical AI across our products, processes and partnerships. These are guiding principles we will use during AI development, distribution or when we’re helping others create AI systems. Privacy-enabled and secure—Designed and used to respect individual’s privacy, be secure and minimize the risk of errors, unintended or malicious use. Human focused—Respect human rights and abide by law throughout their life cycle, designed and used with mechanisms and safeguards, such as capacity for human determination or oversight, to support responsible use and prevent misuse. Inclusive—Designed and used to be inclusive, minimizes harmful bias, ensures fair and equal treatment and access for individuals. Responsible—Designed to be used responsibly and mechanisms should be put in place to ensure accountability. AI systems should disclose information to allow a general understanding of the AI, including how AI can consume resources and influence outcomes. AI driven outcomes should be open to challenge. Robust—Subject to a hazard-based safety engineering approach throughout their lifecycle to build in quality testing and technical safeguards to ensure appropriate function, minimize the risk of misuse, and impact of failure. As we have applied these principles in practice, we’ve uncovered gaps where conventional AI is not meeting our aim of Trustworthy AI. Closing these technology gaps will take engineering and ingenuity. If we are successful in applying these 5 principles and also addressing these shortfalls, the result is Trustworthy AI. Bias: How can a model unveil hidden correlations in the training data that can cause discrimination: Explainability: Ensuring humans can remain in the loop in AI-driven decisions, or how to explain if the system works even when it behaves unexpectedly Trust: Trust rooted in confidence you can measure: Confidence in decisions, or how to increase user’s confidence and make users comfortable with measures of stability and model reliability Robustness: How to understand model failures, detecting and preventing adversarial attacks
• #16: Intent of the slide: AI importance and its adoption Artificial intelligence and machine learning (AI and ML) are no longer the new game in town. As the C-suite continues to focus on shifting technological needs, AI is one of their highest priorities. 98% of customers say they’re already using AI in some capacity or conducting proof-of-concept or pilot programs. Almost half of organizations put AI among their organization’s top 2 initiatives and 2/3 of dedicated AI budgets will continue or increase over the next year. While AI certainly has significant mindshare, the depth, sophistication, and level of understanding of AI varies widely across organizations. It is important to start from Data. A comprehensive data strategy is the foundation for a fully developed AI Strategy. A lack of trust in data impedes further adoption: 63% of orgs still in the early stages of AI/ML development are skeptical of the data they work with. Quick and easy access to Data is key for the success of advanced AI implementations. Indeed, only the 18% of orgs still in the early stages of AI/ML development, find easy and quick the access to their Data. The most AI advanced organizations are scaling AI into Production with specialized tools and expertise, combining them in MLOps frameworks. This is still a minority as we found that 14% have fully realized their AI Strategy and 89% need help scaling AI in Production. Intent of the slide: Key principles for AI, which will be covered in later slides in the deck What do the winning brands know? They all agree that data is the lifeforce that’s essential for activating next-gen operating and business models for their business For some, this is about delivering rich experience across billions of moments For others, it’s about accelerating decision velocity to cut through organizational lethargy And lastly, many are using data to see around the corner and to drive continuous innovation They all agree that Artificial Intelligence success comes driving incremental business value For some, this is about improving decision-making to reduce human error and move faster For others, it’s uncover deeper insights improving quality and faster time to market And lastly, many are using AI to deliver cost efficiency optimizing outcomes and productivity They all agree that ML Ops is needed for bringing AI at Scale combining specialized software and expertise For some, this is about integrate AI into existing IT to automate and collaborate better For others, it’s a moving to Production pilots and standardize practices for optimization and efficiency benefits And lastly, many are using ML Ops to overcome the last mile, operationalizing and generate business value
• #17: Intent of the slide: Closing and Next steps