AI-assisted programming: A revolution in the way we code
In this talk we explore how Large Language Models (LLMs) are trained, how they work and the results they achieve on various benchmarks
We'll then do 3 live demos leveraging specific LLMs:
1) GitHub Copilot X: an advanced AI-pair programmer tool from Microsoft. GitHub Copilot X has been trained on terabytes of source code freely available on GitHub and is capable of many coding tasks: write code from text description, explain code, suggest improvements, fix bugs, write unit tests, do code reviews, etc. We'll demo the various capabilities.
2) Azure OpenAI API: is an API on top of OpenAI models. With carefully crafted prompts we can use GPT3.5 Turbo to solve fairly arbitrary Natural Language Processing tasks, effectively using LLMs as a generic Machine Learning model.
3) GPT-4 Data Analyst: this is a GPT-4 model trained to be an assistant data scientist. It can be fed an arbitrary file and will extract insights from the raw data in a matter of minutes by querying the data in natural language.
08/11/2023 14:00
Multimodal Systems
José Santos holds a Licenciatura em Engenharia Informática (2004), a Masters degree in Artificial Intelligence (2006) both from Faculdade de Ciências e Tecnologia da Universidade Nova de Lisboa and a PhD in Computer Science (Logic-based Machine Learning algorithms) from Imperial College, London (2010). José Santos has been a Software Engineer at Microsoft in the Bing division since 2011, and a Principal Software Engineer since 2020. José is a co-inventor of two Microsoft patents and has worked on multiple Bing teams over the past 12 years: query alterations, autosuggest, shopping relevance, price insights and, more recently, is working on the integration of generative pre-trained (GPT) language models in various product features.
