Summary
Highlights
The video begins with an introduction to the current state of AI, emphasizing its rapid evolution. GPT-4 was a major milestone, but AI continued to advance rapidly in 2025 with the release of GPT-5 and other models, profoundly changing the AI landscape. In the last five months, AI has progressed more than in the past three and a half years since ChatGPT's launch.
AI's remarkable efficiency is highlighted, with a task that once cost €4,500 in computation now costing only €1.2. This massive increase in efficiency (4000x improvement) means that humans cannot compete with AI and must instead collaborate with it. Examples of AI's impact include solving an 80-year-old mathematical problem, diagnosing diseases with 99% accuracy, and improving legal efficiency by being 20% better, 100 times faster, and 99.97% cheaper than senior lawyers. The emergence of powerful AI models like MITOS also poses cybersecurity challenges, detecting thousands of vulnerabilities in systems, leading to global discussions among world leaders and even religious figures.
The course will teach participants how to maximize the use of ChatGPT 5.5, create software using natural language without coding, develop AI agents that can perform tasks, and build a personal AI assistant. An example of a personal AI assistant named Jarvis is demonstrated, capable of conversational interaction and performing various tasks.
The concept of AI agents, which can self-prompt and delegate tasks, is introduced as a major breakthrough. Unlike previous AI automatizations, current AI agents truly function as delegable entities. The example of IKEA's AI agent, Billy, shows how it solved 47% of customer service tasks (3.2 million interactions), a job previously done by 8,500 people. IKEA opted to retrain these employees as retail designers, leading to an additional 1.3 billion euros in revenue, demonstrating the correct symbiosis between AI and human value contribution.
The importance of good prompting is emphasized with the 'Sh*t In, Sh*t Out' principle. Most users (98%) use AI incorrectly. The video categorizes tasks into three types: punctual (one-time), themed (related to a common topic), and repetitive. For punctual tasks, a detailed prompt is crucial, following a five-part formula: context/role, query/task, specifications, quality criteria, and response format. This ensures AI provides optimal results, similar to giving a thorough briefing to a human colleague.
A demonstration shows how to use the detailed prompt formula in ChatGPT to generate a comprehensive financial report from notes and an Excel sheet. The AI successfully analyzes the data and provides an executive abstract, results overview, comparisons, and recommendations in 36 seconds. A trick is revealed where a short command ($p) in customized instructions can automatically expand a brief prompt into the full detailed formula, essentially programming ChatGPT to generate elaborate prompts for optimal results.
ChatGPT can be integrated into Excel as an add-on, allowing users to leverage AI capabilities directly within their spreadsheets. A demonstration shows how to ask ChatGPT to create an Excel sheet for team building costs using natural language. The AI generates a formatted spreadsheet with formulas and even a graph, eliminating the need for users to know complex Excel functions.
For themed tasks (multiple related tasks), ChatGPT's 'Projects' feature is introduced. Projects act as intelligent folders where relevant information (preferences, documents) can be stored once, and AI will use this context for future queries within that project. An example of a 'Vacations' project demonstrates how AI can generate itineraries, identify unbooked hotels, and provide recommendations based on previously input personal preferences and documents. Projects can also be shared, enabling collaborative work.
For repetitive tasks, 'GPTs' (Generative Pre-trained Transformers) are the ideal tool. GPTs are custom AI models trained to perform specific tasks repeatedly according to user-defined instructions. An example is shown where a 'Meeting Pro' GPT summarizes a meeting transcript and analyzes power dynamics, providing detailed insights without needing explicit prompts for each request. Creating a GPT is simplified by providing instructions in natural language, following a specific formula (objective, act as, specific instructions, reinforcement). A GPT is also demonstrated for analyzing complex public tender documents, saving hours of manual work.
The BBVA bank's pilot program, where 3,300 employees created over 3,000 custom GPTs in 5 months, saved 2.8 hours per employee weekly. This led to BBVA implementing ChatGPT licenses for all employees, demonstrating the significant return on investment. The key takeaway emphasizes distinguishing between unique, themed, and repetitive tasks and applying the appropriate AI tool (chat GPT, Projects, or GPTs, respectively) to maximize efficiency. A quick quiz reinforces these concepts, with the audience correctly identifying the right tool for different scenarios.