Explore how RNNs address memory and persistence with loops and a hidden state that updates from the current input and previous state to enable time series forecasting and next-word prediction.

Learn how LSTMs use memory cells and four interacting layers to capture long term dependencies, address vanishing gradients, and improve next word predictions with input, forget, and output gates.

Build a simple lstm language model with keras and tensorflow 2.x, using tokenization, padding, and embeddings to predict next words, then discuss limitations of a tiny dataset and simple architecture.

Transformers enable parallel processing and excel at long-range dependencies via multi-head self-attention with query, key, and value. The lecture shows building a PDF question answering tool with Hugging Face models.

Build a chatgpt-like chatbot in a Streamlit web app using an OpenAI API key, focusing on data privacy while generating topics, sections, and content.

Upload a PDF and chat with contents using Streamlit, PDF plumber, and OpenAI API key; extract text, prompt the model, and display answers, with warnings for missing PDFs or questions.

Build a web app that fetches content from a user-provided page, extracts paragraph text with beautifulsoup, and uses a Hugging Face transformers summarizer via streamlit to display bullet points.

Tune a GPT-2 model on your dataset to improve privacy and knowledge repository. Use Colab, PyTorch, and Transformers to train a custom dataset with padding for roleplay prompts.

Fine-tune the GPT-3.5 model with a five-step workflow, prepare json-lines data with roles like system, user, and assistant, and deploy a ready-to-use fine-tuned model.

Explore LangChain's modular components, chains, and agents that streamline language model applications and enable interactions with OpenAI or Hugging Face, with practical setup and testing.

Explore prompt engineering with long chain and Lang, using prompt templates to reduce bias and craft structured or unstructured responses, demonstrated by extracting NPS content from a PDF.

Employ a csv agent powered by OpenAI and pandas to answer natural language questions about an insurance dataset, revealing record counts and BMI-based obesity insights.

Explore evaluation metrics for llms: perplexity, bleu, and rouge. Learn about n-grams, brevity penalties, and bias detection with the weak method and cosine similarity in embeddings.

Predict the next word with a GPT-2 model and compute perplexity to evaluate language models. Use transformers and torch, with tokenizers, and run on GPU or CPU depending on availability.

Evaluate summarization quality with Rouge metrics by comparing generated and reference summaries, using Rouge one, Rouge two, and Rouge L, with stemming and reporting precision, recall, and F1.

Assess bias in Bert using a word embedding association test with male/female names and career/family words, computing embeddings, cosine similarity, and a wheat score; p value 0.0885 signals borderline significance.

Compare retrieval augmented generation (rag) with fine tuning, showing how rag retrieves external knowledge to reduce hallucinations and improve accuracy in dynamic knowledge intensive tasks.

Explore recent advancements in RAG, from naive to modular frameworks, with pre and post retrieval, iterative and recursive retrieval, and modular racks that integrate search, memory, and adapters.

Build a retrieval augmented generation rag chat app for pdfs by extracting text from a document, retrieving external information from Wikipedia, and generating responses with OpenAI.

Implement a retrieval augmented generation system using a sentence transformer to fetch the relevant FAQ answer from a similarity index and generate a detailed response with GPT two language model.

Compare RAG systems, which merge retrieval with generative AI to deliver accurate outputs from external sources, with AI agents that autonomously plan actions and adapt to dynamic goals.

Build a financial analyst agent using the five framework with OpenAI's GPT-4 and Yahoo Finance tools. Access real-time stock prices, company information, analyst recommendations, and news within multimodal workflows.

Develop an ai powered management consultant agent to curate engaging, high quality Forbes article content using GPT-4 and XR tools, emphasizing research, citation, and structured writing.

Create an interactive ai powered data analysis assistant that loads local datasets, validates file paths, and answers questions with natural language using gpt-4 and duckdb.

Develop a multifunctional AI assistant with a web agent and a finance agent, powered by the five framework and GPT-4, for web searches and financial data insights.