GenAI for clients No pain, no gain
summary
Fine-tuning produces a custom LLM that turns a 70‑year, multi‑sheet Excel into charts by first cleaning misspellings, mapping historical name changes (Google→Alphabet, Facebook→Meta), and adding column descriptions; training runs 24–27 iterations across Llama2/Lama three/Mistral variants and yields 95% accuracy on a 1k test set. Charts are validated with a pixel-by-pixel CV comparator; visual pipelines use zero-shot Segment Anything plus a 12 Labs index for sketch→video similarity to find character frames and support piracy protection.
AI-written summary · about
chapters (11)
- 0:00🚀 Introduction: The Challenges of Working with GenAI for Clients
- 2:30The Project Overview: Fine-tuning Custom LLMs for Business Data
- 5:00Data Cleaning: Handling Inconsistent Historical Data
- 8:00Working with Human Language: Creating User-Friendly AI Responses
- 11:30LLM Limitations and the Importance of Proper Prompting
- 15:00Creating Accurate Charts with GenAI and Improving Model Accuracy
- 18:00The Role of AI in Content Protection: Working with Animation and Cartoons
- 22:00Visual Language Models for Character and Object Recognition
- 26:00Addressing Client Concerns: Managing AI Expectations and Challenges
- 30:00Zero-Shot Training and Fine-Tuning for Complex Data Requirements
- 33:00Conclusion: Real-World Applications and Future Directions for GenAI
related talks
27mAI Demystified Unveiling the Intelligence Behind the Machine
48mPracticalities of Fine-Tuning Llama 2 with AI Studio
48mMastering the AI Shift: Adaptive Strategies for Software Evolution
41mAn Art, a Science, or Your Next Job Title?
29mGenerative-AI: Usecase-Specific Evaluation of LLM-powered Applications
49mGenAI for Business on the Modern Data Stack
47mBuilding and Testing AI-Agent Powered LLM Applications: A Live Demonstration
49mNext-Gen Documentation with AI