01.-Ingeniería de IA desde cero: fundamentos, herramientas y diseño de aplicaciones.

The New Scale of Artificial Intelligence

The Impact of Scale on AI Development

• The term "scale" defines the evolution of artificial intelligence post-2020, highlighting its transformative impact.

• Large models like Chat GPT and Gemini consume significant energy, contributing to a notable fraction of global energy consumption.

• A paradigm shift has occurred with the emergence of AI engineering, making advanced models accessible as services via APIs without requiring deep expertise in machine learning.

Historical Context and Evolution

• The roots of AI trace back to the 1950s with pioneers like Claude Shannon modeling language statistically.

• The development of large language models (LLMs), such as GPT2 and GPT3, was enabled by self-supervision techniques that allow models to learn from vast amounts of text data available online.

Advancements in Model Capabilities

• Recent advancements have led to foundational models that integrate multiple modalities beyond text, including images and audio.

• These foundational models serve as versatile bases for various applications rather than being limited to single tasks.

Strategies for Adapting Models

• Three key strategies exist for adapting foundational models: prompt engineering, retrieval-Augmented Generation (RAG), and fine-tuning.

• Prompt engineering involves giving precise instructions to guide model responses.

• RAG combines model outputs with external information sources for enhanced accuracy.

• Fine-tuning allows partial retraining with specific examples for tailored performance.

Growth of AI Engineering Profession

• The role of AI engineers is emerging rapidly; they adapt existing models into practical solutions rather than creating them from scratch.

• Factors driving this growth include increasing model capabilities, massive investments from companies/governments, and accessibility allowing anyone with an idea to create functional applications.

A New Era in Technology

Transformative Potential of Foundational Models

• We are witnessing a pivotal moment in technology where foundational AI models transition from experimental tools to widely used resources across industries.

Integration into Daily Life

• With simple prompts or API calls, users can now build applications capable of understanding text or generating multimedia content without extensive coding knowledge.

Explosion of Use Cases

• Thousands of new AI tools are launched monthly; while many are variations, some represent genuine innovations that redefine possibilities within various sectors.

The Impact of AI on Various Industries

Coding Assistance

• AI tools like GitHub Copilot, Cursor, and GPT Engineer are revolutionizing software development by enabling faster coding, automatic test generation, and even creating entire websites from sketches.

• The integration of AI in programming not only saves time but also transforms learning methods, design processes, and debugging techniques.

• Developers using AI assistants report productivity increases of 20% to 50%, depending on the task type.

Visual Creativity

• Applications such as Mid Journey and Pica Labs allow users to generate images and videos from text prompts, automating creative processes in visual media.

• In professional settings like advertising and design, companies leverage these tools for cost savings and enhanced creative speed.

Writing and Communication

• AI has become commonplace in writing tasks including emails, reports, social media posts; it helps with error correction and style improvement while handling repetitive tasks.

• Many professionals utilize tools like ChatGPT for drafting important communications or SEO content across various corporate departments.

Personalized Education

• In education, AI is changing how students learn by providing personalized experiences that adapt to individual learning styles and preferences.

• Tools such as virtual assistants can explain concepts or create exercises without replacing teachers but rather enhancing their ability to guide students effectively.

Conversational Bots

• Conversational bots are increasingly sophisticated in customer service roles; they can respond accurately via text or voice interactions.

• These bots offer 24/7 availability for user inquiries while also evolving through each interaction to improve responses over time.

Information Aggregation

• With vast amounts of data available today, AI assists in summarizing information efficiently—extracting key points from documents or messages.

• Tools like Perplexity help researchers access relevant information quickly for academic research or project management purposes.

Workflow Automation

• Intelligent automation powered by AI streamlines workflows by coordinating tasks across different platforms while making simple decisions autonomously.

• This adaptability allows businesses to enhance operational efficiency significantly as many existing automation processes now benefit from contextual improvements.

Industry Adoption Trends

• Over 70% of companies are exploring generative AI applications; one-third actively use them internally across various sectors including technology, finance, education, and more.

• New job roles are emerging alongside traditional ones transforming due to the influence of AI technologies. Ethical considerations accompany this rapid evolution.

Evaluating AI Application Development

The Potential of Artificial Intelligence

• La inteligencia artificial (IA) ofrece un amplio rango de posibilidades que antes parecían ciencia ficción, ahora realizables con poco código y modelos fundacionales.

• Antes de construir una aplicación, es crucial preguntarse si realmente se debe hacer, ya que no todo lo que es posible debe ser realizado.

Deciding to Build an Application

• La decisión de construir comienza con entender el propósito: ¿se responde a una necesidad real o se busca explorar nuevas oportunidades?

• Existen tres escenarios para considerar la construcción:

• Urgente: Actuar es necesario para no quedarse atrás frente a competidores.

• Ventaja: Implementar IA puede mejorar procesos y reducir costos.

• Exploración: Experimentar con IA sin comprometerse completamente.

Build or Buy?

• Una vez decidido construir, surge la pregunta de si desarrollar internamente o utilizar soluciones existentes.

• Si la aplicación es central al negocio, puede valer la pena desarrollarla a medida; si es complementaria, comprar puede ser más eficiente.

Types of AI Applications

• Las aplicaciones pueden clasificarse en tres dimensiones:

• Reactiva vs. Proactiva: Responde a solicitudes del usuario o actúa por iniciativa propia.

• Estática vs. Dinámica: Cambia con el tiempo o permanece constante para todos los usuarios.

• Complemento vs. Elemento Central: Funciona sin IA o depende completamente de ella.

Human Interaction in AI Applications

• La interacción humana es clave en aplicaciones de IA; se debe definir cuándo interviene un humano en el proceso decisional.

• Se sugiere un modelo de tres fases (craw, walk, run):

• Comenzar con humanos evaluando y ajustando,

• Permitir interacciones controladas entre IA y equipos internos,

• Finalmente permitir mayor autonomía a la IA bajo métricas claras.

Advantages and Risks of Developing AI Applications

• Desarrollar aplicaciones de IA tiene potencial pero también riesgos como la obsolescencia rápida debido a avances tecnológicos constantes.

• Es importante identificar ventajas competitivas como tecnología propia, datos únicos y distribución efectiva para diferenciarse en el mercado.

Measuring Success and Maintenance

• Definir cómo medir el éxito antes del lanzamiento es esencial; considerar métricas como calidad de respuestas y satisfacción del usuario.

• Planificar no solo cómo iniciar sino también cómo mantener actualizada la aplicación ante cambios rápidos en tecnología y regulaciones.

The Technological Stack of AI Engineering

Overview of AI Infrastructure

• The rise of artificial intelligence (AI) in daily life is supported by a complex and well-designed infrastructure, emphasizing the need for an entire ecosystem beyond just having a good model.

Layers of AI Application Development

• Building applications based on foundational models involves three distinct layers, each with its own tools, challenges, and responsibilities.

1. Application Development Layer

• This visible layer connects directly with users, focusing on interface design, prompt definition, user flow construction, and final experience evaluation. It has seen significant innovation recently and is accessible to new creators.

2. Model Development Layer

• A more technical layer where teams create, train, or fine-tune models. Key tasks include data curation, annotation, performance testing, and traditional machine learning engineering practices.

3. Infrastructure Layer

• The foundational layer that includes hardware management and resource allocation necessary for large-scale model operation. A well-designed infrastructure differentiates between demos and real production products.

Current Trends in AI Ecosystem

• An analysis from March 2024 revealed explosive growth in application development tools within the AI sector; these tools now surpass established frameworks like React or Vue in popularity.

Principles Sustaining AI Engineering

• Despite evolving tools, core principles remain unchanged: continuous experimentation (testing and adjusting), user feedback (listening to real data), and alignment with business objectives (ensuring value addition).

Distinction Between ML Engineering and AI Engineering

• Machine Learning (ML) engineering focuses on creating models from scratch while AI engineering adapts pre-trained models to specific problems—this shift impacts scale requirements and evaluation methods significantly.

Adaptation Techniques in AI Engineering

• Adapting models can involve simple adjustments like prompt engineering or deeper modifications through fine-tuning; both techniques have their appropriate contexts for use.

Prompt Engineering vs Fine-Tuning

• Prompt Engineering: Involves providing better instructions without altering the model itself; it's easy to implement and cost-effective.

• Fine-Tuning: Modifies model weights for deeper adaptation but requires more resources; knowing when to apply each technique is crucial.

Conclusion on the Evolution of Product Design with AI

• The structured approach to AI—application development, model adaptation, infrastructure—is essential for making intelligent systems useful at scale while adhering to solid principles of experimentation and responsiveness.

New Professional Profiles Emerging

• The evolution brought about by foundational models is not only changing product design but also leading to new professional roles within teams focused on integrating AI into digital products effectively.

Emergence of AI Engineers

The Role of AI Engineer

• The term "AI Engineer" has become increasingly prevalent in job listings, reflecting its growing importance in the tech industry.

• An AI Engineer builds functional applications on existing foundational models, requiring an understanding of model behavior and integration rather than training from scratch.

• This role merges software engineering with AI thinking, bridging traditional development with artificial intelligence.

Evolution of Workflows

• Traditional Machine Learning workflows followed a clear sequence: data collection, model training, and product development.

• In contrast, modern AI workflows reverse this order to product-first: prototypes are built using pre-trained models before gathering real user data for adjustments.

• This agile approach allows for rapid iteration based on direct user feedback and behavior analysis.

Importance of User Interfaces

• In the new landscape, interfaces play a crucial role; it's not just about having a powerful model but also how it is presented and made accessible to users.

• Various interaction methods are emerging beyond text-based chat interfaces, including voice assistants and immersive environments like AR/VR.

Feedback Mechanisms

• Modern systems can capture direct user interactions and feedback more effectively than traditional metrics like clicks or forms.

• This shift leads to shorter feedback loops that enhance both product improvement and model adjustment.

Summary of Changes in AI Development

• The transition from laboratory-focused AI to integral product development signifies a shift towards prioritizing user experience over mere data-centric approaches.

• Future discussions will explore applying these principles throughout the entire lifecycle of an AI application.