Zero-to-production: bootstrapping text-based AI applications with huge language models
Large language models, such as AI21 Labs’ Jurassic-1 and OpenAI’s GPT-3, exhibit unprecedented versatility in performing many different language tasks.
Using prompt engineering to adapt the model’s behavior to their specific needs, developers and even non-coders can build sophisticated AI applications from scratch in a matter of days. In this talk, we review the technology powering large language models and discuss their capabilities and limitations.
A significant challenge that developers relying on large language models face is scaling up their usage economically without sacrificing quality. We will present approaches to overcome the cost-quality tradeoff in the context of a product life cycle, by starting with rapid prototyping using prompt engineering and large language models, and then evolving to training and deploying a more affordable custom model, optimized to perform a specific task at scale.
VP Platform at AI21 Labs
Dan Padnos is VP of Platform at AI21 Labs, a Tel Aviv-based startup on a mission to transform the way humans read and write. Dan leads AI21 Studio, a developer platform providing access to the company’s state-of-the-art Jurassic-1 language models.
Prior to this, he led algorithm development for Wordtune, AI21 Labs’ premium writing assistant, and served in the IDF as a researcher and algorithm developer. In his previous life as a physics MSc student, Dan studied the flow of gas in ancient galaxies using large-scale computer simulations.