Back

3 levels of human involvement in AI systems. I’ve explained each approach in simple steps below. 𝗛𝗜𝗧𝗟 (𝗛𝘂𝗺𝗮𝗻-𝗶𝗻-𝘁𝗵𝗲-𝗟𝗼𝗼𝗽) Humans are actively involved at every step, from collecting data to monitoring results and improving the model. 1. Gather information and define clear goals. 2. Record logs to track every action and system event. 3. Measure key metrics like accuracy and performance. 4. Clean and prepare the collected data for analysis. 5. Pick the right tools and frameworks for the task. 6. Normalize the data so the model reads it correctly. 7. Build and train models using curated data. 8. Detect anomalies or unusual behavior. 9. Trigger automated actions for smaller issues. 10. Deploy the system and analyze its real-world results. 11. Continuously optimize with human feedback and new data. ✅ 𝗜𝗻 𝘀𝗵𝗼𝗿𝘁: Humans guide and monitor the AI throughout the process. ___________________________________ 𝗛𝗢𝗧𝗟 (𝗛𝘂𝗺𝗮𝗻-𝗼𝗻-𝘁𝗵𝗲-𝗟𝗼𝗼𝗽) Humans supervise and fine-tune the system instead of being part of every decision. 1. Choose the right large language model for your goal. 2. Define what specific task the model should complete. 3. Prepare and organize data for testing or fine-tuning. 4. Use open-source and external resources for flexibility. 5. Integrate the LLM with APIs and tools. 6. Combine systems so everything works seamlessly. 7. Test and validate output quality. 8. Design and refine prompts to improve accuracy. 9. Fine-tune the model to perform consistently well. 10. Measure accuracy and monitor real-time results. 11. Track model performance over time. 12. Iterate and improve based on ongoing feedback. ✅ 𝗜𝗻 𝘀𝗵𝗼𝗿𝘁: Humans oversee, guide, and correct the system when needed. ___________________________________ 𝗛𝗙𝗢𝗧𝗟 (𝗛𝘂𝗺𝗮𝗻-𝗼𝗳𝗳-𝘁𝗵𝗲-𝗟𝗼𝗼𝗽) Humans design the system once, and it runs automatically, requiring little to no manual supervision. 1. Collect large volumes of raw data. 2. Clearly define the main problem to solve. 3. Process and structure the information for training. 4. Organize data into clean, usable formats. 5. Manage both structured and unstructured data types. 6. Select the best algorithm for the task. 7. Train and refine models for top accuracy. 8. Clean and polish datasets throughout training. 9. Engineer useful features that improve predictions. 10. Deploy the model into production. 11. Monitor automatically and retrain when needed. 12. Scale and automate the entire process end-to-end. ✅ 𝗜𝗻 𝘀𝗵𝗼𝗿𝘁: The system runs on full automation, minimal human involvement. 𝗜𝗻 𝘀𝗵𝗼𝗿𝘁: • 𝗛𝗜𝗧𝗟 = Human works 𝘸𝘪𝘵𝘩 AI at every stage. • 𝗛𝗢𝗧𝗟 = Human 𝘴𝘶𝘱𝘦𝘳𝘷𝘪𝘴𝘦𝘴 AI and steps in when required. • 𝗛𝗙𝗢𝗧𝗟 = AI runs 𝘪𝘯𝘥𝘦𝘱𝘦𝘯𝘥𝘦𝘯𝘵𝘭𝘺 with full automation. You can apply this framework to build AI systems with the right human control. ✅ 𝗥𝗲𝗽𝗼𝘀𝘁 𝗳𝗼𝗿 𝗼𝘁𝗵𝗲𝗿𝘀 𝗶𝗻 𝘆𝗼𝘂𝗿 𝗻𝗲𝘁𝘄𝗼𝗿𝗸 𝘄𝗵𝗼 𝗰𝗮𝗻 𝗯𝗲𝗻𝗲𝗳𝗶𝘁 𝗳𝗿𝗼𝗺 𝘁𝗵𝗶𝘀.