Generative AI

Responses

1. 

   Winner Iwinosa Ehizode June 27, 2025

   What I Learned from the Generative AI Full Course

   I watched the “Generative AI Full Course-Gemini Pro, OpenAI, Llama, Langchain, Pinecone…” video as required. It was much more extensive than I expected, over 30 hours of instruction. Working through it taught me a lot, and here’s a summary of my main takeaways:

   1. Foundations of Generative AI
   • I learned the difference between generative and discriminative models. Generative models (like VAEs and GANs) can create new content, while discriminative models classify existing data.
   • I gained a strong understanding of variational autoencoders (VAEs) and how they learn efficient representations by encoding and reconstructing data.
   • I also got into GANs (Generative Adversarial Networks), how a generator and discriminator compete to generate realistic images, and how advanced variations improve output quality and stability.

   2. Transformer Models & LLMs
   • The course dived deep into Transformer architecture, explaining self attention, positional encoding, and how these models process sequence data.
   • I learned how GPT style models (like OpenAI’s GPT) are pre-trained on massive text corpora and later fine tuned for tasks like summarization, Q&A, or creative writing.

   3. Using OpenAI, Gemini Pro, and Llama in Practice
   • I practiced using the OpenAI API: sending prompts, managing tokens, and interpreting results.
   • I got introduced to Google’s Gemini Pro, a powerful multimodal LLM, and learned how it compares to GPT-like models.
   • I also explored Meta’s Llama models, experimenting with lightweight yet powerful open source alternatives.

   4. Building with LangChain
   • I discovered the LangChain framework, which helps build pipelines that chain prompts, memory, retrieval, and response generation.
   • This included using memory components, structuring Q&A chains, and using agents to automate reasoning steps.

   5. Vector Databases & Retrieval
   • I learned how embeddings convert text into numerical vectors representing meaning.
   • I saw how to store and retrieve embeddings using vector databases like Pinecone and ChromaDB.
   • I built retrieval augmented generation (RAG) systems, combining retrieved documents with LLMs to answer queries knowledgeably.

   6. Real World Projects
   • The instructors walked through creating a medical chatbot (complete with retrieval and memory).
   • Another project used Gemini Pro for content creation and response capabilities.
   • I learned how to deploy these systems and integrate them into applications, understanding dependencies, logging, and API calls.

   7. Ethical and Future Perspectives
   • The course briefly touched on responsible AI, including concerns like deepfakes, copyright, and job displacement.
   • I also explored emerging trends agents AI, multimodal models, and the impact of projects like Gemini Pro.

   Overall, I started with a basic understanding of AI and ended up being able to:
   1. Explain how VAEs, GANs, and Transformers work.
   2. Use APIs from OpenAI, Gemini Pro, and Llama to generate and refine content.
   3. Build end-to-end AI systems with LangChain, memory, embedding retrieval, and vector DBs like Pinecone.
   4. Reflect on AI’s world impact and responsible use.

   This course transformed my overnight goal into a long term journey, but I emerged with both theoretical knowledge and practical skills I can apply immediately. Thank you for the opportunity to learn and grow through this scholarship.