Generative AI

Introduction to Generative AI course

Lambert Fatoux

Introduction to the class

About me

  • Lambert, master’s degree in computational mathematics at ENPC and ENS Paris-Saclay
  • Companies I’ve worked for: MétéoFrance, Phillips, Ministry of Defense
  • 3 years of teaching in engineering schools on work related topics: MLOps, Generative AI, Python best practices
  • If you have any questions, feel free to contact me on LinkedIn or by mail at [email protected]
  • You may find my personal github for all the resources about this course and others: lambzcode

During this class

  • Ask questions, be curious !
  • Things might move very quickly, we’ll try to keep up the pace so that you have state-of-the-art picture of Generative AI
  • I’ll try to share with you the latest AI news if any, at the end of these slides you’ll find some ressources that I use to stay up to date

During this class

  • Continuous assessment 20% - Participation 10% - Final project 70% ​
  • Group Project Guidelines: teams are expected to create a small-scale application leveraging LangGraph and LangChain, combined with OpenAI’s LLM, to solve a particular data-related problem. ​
  • Example: developing an AI-powered customer service chatbot that utilizes LLMs for intelligent responses and LangChain to control the conversation flow. ​
  • By the next session and starting at this course’s conclusion -> select a team and a project

Outline

  1. Introduction
  2. Transformers
  3. Pretraining
  4. Reinforcement learning
  5. Advanced Prompting & structured outputs
  6. RAGs
  7. Agents & tools to use
  8. Diffusion mechanisms
  9. Evaluation harnesses

From AI to ChatGPT

Before talking about generative, what even is AI ?

What are ML, AI and DL ?

  • Artificial Intelligence (AI): Any system of imitation of human thought and actions

  • Machine Learning (ML): Algorithm whose performance improves with new data

  • Deep Learning (DL): Learning algorithms based on artificial neural networks

From AI to ChatGPT

Before talking about generative, what even is AI ?

Machine learning timeline

From AI to ChatGPT

What is NLP ?

NLP, or Natural Language Processing, is an innovative approach to subdomain interdisciplinary of linguistics and artificial intelligence which focuses on the interaction between computers and human language, in particular on how to program computers so that they process and analyze large quantities of data in natural language.

The aim is to create models capable to “understand” the content of documents, including the contextual nuances of the language they contain

NLP common tasks

  • Sentence boundary disambiguation
  • Language detection
  • Spelling correction
  • word segmentation
  • part of speech tagging
  • chunking
  • parsing
  • representation
  • text classification
  • Sentiment analysis
  • Named Entity Recognition
  • Relationship extraction
  • Coreference resolution
  • Slotfilling
  • Entity linking
  • Knowloedge representation and reasoning
  • Semantics
  • Semantic role labeling
  • Textual entailment
  • Question answering
  • Language generation
  • Machine translation
  • Automatic summaraization
  • Automatic speech recognition
  • Text to speech
  • Dialog Systems and Chatbots
  • Optical Text Recognition
  • …

Focus on NLP

Generative AI

  • Generative AI uses existing content to learn to generate new one.

  • This makes it possible to generate figures/images, videos, text, code, voice, music etc

Ressources