Learning Library

← Back to Library

Rocket Launch Analogy for AI Training

Key Points

  • Training large language models is likened to launching a rocket: it demands massive compute resources, months of effort, and meticulous planning because once training starts, design changes aren’t possible.
  • Kate Soule, acting as “mission control” at IBM, emphasizes that her business‑strategy background drives a focus on ensuring LLM research delivers real, tangible value for clients rather than just technical breakthroughs.
  • Generative AI extends traditional AI by not only analyzing data but also creating new content, enabling use cases such as automated customer service, code generation, and complex document extraction that boost productivity and cut costs.
  • Foundation models are large, general‑purpose systems trained on vast unlabeled data via unsupervised learning, which can then be fine‑tuned for a wide variety of applications, making them more versatile than task‑specific traditional ML models.
  • Building internal expertise and dedicated teams now is crucial, as generative AI is becoming a key business differentiator and organizations need the capability to innovate and shape the evolving AI landscape.

Full Transcript

# Rocket Launch Analogy for AI Training **Source:** [https://www.youtube.com/watch?v=1JzMSbcInxc](https://www.youtube.com/watch?v=1JzMSbcInxc) **Duration:** 00:08:14 ## Summary - Training large language models is likened to launching a rocket: it demands massive compute resources, months of effort, and meticulous planning because once training starts, design changes aren’t possible. - Kate Soule, acting as “mission control” at IBM, emphasizes that her business‑strategy background drives a focus on ensuring LLM research delivers real, tangible value for clients rather than just technical breakthroughs. - Generative AI extends traditional AI by not only analyzing data but also creating new content, enabling use cases such as automated customer service, code generation, and complex document extraction that boost productivity and cut costs. - Foundation models are large, general‑purpose systems trained on vast unlabeled data via unsupervised learning, which can then be fine‑tuned for a wide variety of applications, making them more versatile than task‑specific traditional ML models. - Building internal expertise and dedicated teams now is crucial, as generative AI is becoming a key business differentiator and organizations need the capability to innovate and shape the evolving AI landscape. ## Sections - [00:00:00](https://www.youtube.com/watch?v=1JzMSbcInxc&t=0s) **LLM Training as Rocket Launch** - The speaker likens large language model training to a resource‑intensive rocket launch, emphasizing meticulous planning, the inability to tweak once training begins, and a business‑focused mission to deliver real, tangible value. - [00:03:17](https://www.youtube.com/watch?v=1JzMSbcInxc&t=197s) **Framework for Responsible GenAI Adoption** - The speaker outlines a step‑by‑step approach—building a skilled team, piloting a low‑risk use case, defining value and compliance requirements, ensuring transparent and trustworthy operations, and selecting proper evaluation metrics—to successfully integrate generative AI into business. - [00:06:24](https://www.youtube.com/watch?v=1JzMSbcInxc&t=384s) **Continuous Model Retraining & Governance** - The speaker stresses IBM’s practice of regularly retraining foundation models to incorporate new data, regulatory updates, and risk‑management best practices, emphasizing the need for a robust AI platform with governance tools that can guide organizations from experimentation to self‑managed deployment. ## Full Transcript
0:00Training. 0:01A new large language model is a bit like launching a rocket. 0:05Five. 0:05It's exciting. 0:06Four. 0:07It's resource intensive. 0:08Three. 0:09It requires an enormous amount of compute power. 0:12Two. 0:13And the training process takes months. 0:15One. 0:16So you need intensive planning and preparation to make sure 0:18you've got the latest and best technologies in place. 0:21Because once you press go and the GPUs fire 0:24up and start training, the rocket has liftoff. 0:27You can no longer tweak the design. 0:29Any new innovation has to wait until the next launch. 0:32And just like rocket launches change the frontier of science, 0:36large language models and the broader class of generative 0:40AI that they belong to called foundation models represent 0:44a paradigm shift in how the world is going to leverage AI. 0:47Zero. 0:48All engines running. 1:03Welcome to AI Academy. 1:05I'm Kate Soule, Senior Manager of Business Strategy 1:08at IBM Research and the MIT-IBM Watson AI Lab. 1:12And in that analogy, I work at mission control. 1:15My job is to oversee at a program level the training and development 1:19of all the large language models for IBM's AI and data platform. 1:23And I come at that role 1:24from a business and consulting background rather than a pure technical one. 1:29But it means that I approach my job and the work that we do 1:33with a focus on trying to make sure that our research has impact on the world, 1:38that what we're doing is solving real business problems 1:41and generating real, tangible value for our clients. 1:45And in terms of real value, the opportunities with generative 1:48AI are extraordinary. 1:51While traditional AI can analyze data and tell you what it sees, 1:54generative AI can use that same data to create something new. 1:58And that's a vital tool for businesses to have because that same power 2:02can be applied to customer service and support, code generation for developers, 2:08extracting key information from complex documents. 2:11More use cases are being developed every day. 2:15Companies can increase 2:16productivity, reduce costs and open up new lines of business, while 2:21traditional machine learning is narrowly focused and purpose-built 2:25for a specific task and takes a lot of human intervention. 2:28Foundation models are bigger, broader general purpose models 2:33that benefit from unsupervised learning, 2:36which means they can be trained on large, unlabeled data sets. 2:40And then afterwards, this general purpose model 2:43can be further tailored for an array of applications. 2:46The types of things these models can do is evolving incredibly quickly. 2:52So now is the time to start building your expertise. 2:55As generative AI becomes a business differentiator, 2:58you're going to want the ability to innovate so that you're not just 3:01following what other companies have done, and you're going to want to be part 3:05of the broader conversation about what AI 3:07is and where the field is going in building that muscle mass 3:11in your organization for how to build and experiment with generative AI. 3:19When looking to get 3:20started, building expertise is critical. 3:24First, you need to establish a team of people 3:26who can become comfortable and fluent working with foundation models 3:30so that they can experiment, testing out new models as they become available, 3:35prototyping on example use cases and so on. 3:38The second step is to pick an internal low 3:41risk use case that you can use as a testing ground. 3:45You could build a prototype and test out deployment. 3:47Then use what you learn as your team gains more experience. 3:51Third, you need to have an in-depth conversation about what 3:55you require to get those real value drivers and revenue drivers. 3:59That generative AI can help you unlock. 4:02For example, you need to determine what requirements around trustworthiness 4:07and other regulatory issues your models need to meet to be deployed in production. 4:12And all those questions only become more relevant as you leave the experimentation 4:17phase and get into the actual building of a model 4:20for real on an application that can drive business impact. 4:24And finally, you need to be able to operate 4:27with a level of responsibility and transparency. 4:30You've got to be transparent regarding data collection, showing 4:33what is and isn't in your data and how it all gets filtered and managed. 4:38You need to be 4:38able to explain how your AI is making decisions. 4:41You want it to be fair and trustworthy 4:44and ready for compliance with upcoming regulations. 4:47The number one success factor in each of these steps 4:51is choosing the right evaluation metrics 4:53that reflect your business tasks and measure the model's robustness, 4:57fairness, scalability and cost for deployment across your business. 5:02And even though that evaluation can be quite difficult for generative AI, 5:06when the right answer could be subjective, when you evaluate across 5:10all these dimensions, you may find that some use cases don't 5:14justify the cost or risk of leveraging a huge model on the cloud. 5:18That's why one model doesn't have to rule them all. 5:23Within IBM research, we are seeing that smaller specialized models. 5:28Now, when I say smaller, I'm 5:29still talking billions, not trillions of parameters and size 5:33can be as proficient as those giant trillion 5:36plus language models when they are evaluated on specialized tasks. 5:41These smaller models are significantly more cost efficient 5:44and can be run more easily on prem to reduce your deployment 5:48risk. 5:53When you're getting started 5:54on your journey with generative AI and looking at all the options available. 5:58My recommendation is to start simple, start with a pre-trained model 6:03and try to do light customizations with your own data 6:05through a process called tuning. 6:07This way you can tailor the model for your specific use cases 6:10while taking advantage of the large 6:12general purpose capabilities that other providers have developed. 6:16It's important, though, to update those pre-trained models every couple of months. 6:21Going back to the rocket ship analogy. 6:24IBM Research has a regular launch cadence, 6:27retraining all of our foundation models multiple times a year 6:31as more information is made available in the world continues to progress. 6:35We want our models to be able to reflect changes. 6:38We also want to make sure that our models consider the latest regulatory 6:42guidance and risk management best practices. 6:45The field and the regulatory guidance around it is constantly evolving. 6:50So models that aren't regularly retrained 6:53with the latest best practices will quickly become stale. 6:56That's why the right AI and data platform is so important. 7:00You should look for a platform that has proven expertise in foundation 7:03models, the governance tools in place 7:07to help you address potential ethical concerns 7:10and can help you transition from experimentation to deployment. 7:14Then, as you get better and more confident over time and training 7:17and owning the models, you'll eventually be able to maintain 7:20and build them out on your own. 7:22There's a lot of complexity to AI in foundation models, 7:25but working through all that complexity 7:28truly is worth it for where it's going to take us, 7:31both in terms of our business successes and our progress as a society. 7:36Think about those NASA scientists and engineers. 7:39Doing something new is never easy, but because they did the work, 7:43we've set foot on the moon and sent probes beyond our solar system. 7:48We can now explore our universe. 7:50Generative AI may not literally be a rocket, but it will help us do more 7:55to travel farther and faster to unlock new possibilities and explore new frontiers. 8:01And I'm so excited to see where it will take us. 8:05Thank you for watching. 8:06Please join us again for more episodes of AI Academy 8:09as we explore some of the most important topics in AI for business.