Large transformers such as GPT-2 and T5 are revolutionizing neural text generation by offering language models with significantly lower perplexity.
However, there are challenges in directly applying these models to drive business value. For example, in sales conversations, call center agents struggle to optimize what they say to convert visitors and drive revenue.
To build a dialogue system that suggests optimal responses and boosts agent performance, we need to generate text that is “action-oriented” -- where an action captures the intent of a message and a good sequence of actions can orient a conversation towards a better outcome.
In this talk, Tim will present how his team combines state-of-the-art text generation with classical dialogue principles to build a production-grade NLP system that drives $100MM incremental revenue for Fortune 500 companies.
Tim is the CTO and co-founder of Cresta (https://cresta.com/), a16z backed company. His work at Stanford as a PhD student and at OpenAI formed the basis for Cresta.
Tim has published work in the leading machine learning conferences and journals including NeurIPS, ICML and the Journal of Machine Learning Research.
Join the Bugout Slack dev community to connect with fellow data scientists, engineers, and ML practitioners. Find new job opportunities/hire engineers, get technical advice, and more: https://join.slack.com/t/bugout-dev/shared_invite/zt-fhepyt87-5XcJLy0iu702SO_hMFKNhQ