Farhad Rahimi

Researcher

What will you create with it?

Generative AI has already revolutionized the world and it’s not slowing down. As a trained computer scientist, if you want to contribute to the revolution of Generative AI, and make an immediate impact in your organization, now is the time to enhance your expertise.  Generative Artificial Intelligence (AI) represents the pinnacle of AI technology, pushing the boundaries of what was once thought impossible. Generative AI models have several capabilities.

For example, to be able to use an LLM model, we give a Prompt question as input and it gives us a Completion as output.

Farhad Rahimi

Researcher

Introduction to Generative AI and Large Language Models

Generative AI is artificial intelligence capable of generating text, images and other types of content. What makes it a fantastic technology is that it democratizes AI, anyone can use it with as little as a text prompt, a sentence written in a natural language. There's no need for you to learn a language like Java or SQL to accomplish something worthwhile, all you need is to use your language, state what you want and out comes a suggestion from an AI model. The applications and impact for this are huge, you write or understand reports, write applications and much more, all in seconds.

• What generative AI is and how Large Language Models work.
• How you can leverage large language models for different use cases, with a focus on education scenarios.

Farhad Rahimi

Researcher

A statistical approach to AI: Machine Learning

A turning point arrived during the 90s, with the application of a statistical approach to text analysis. This led to the development of new algorithms – known as machine learning – capable of learning patterns from data without being explicitly programmed. This approach allows machines to simulate human language understanding: a statistical model is trained on text-label pairings, enabling the model to classify unknown input text with a pre-defined label representing the intention of the message.

Farhad Rahimi

Researcher

Neural networks and modern virtual assistants

In recent years, the technological evolution of hardware, capable of handling larger amounts of data and more complex computations, encouraged research in AI, leading to the development of advanced machine learning algorithms known as neural networks or deep learning algorithms.

Neural networks (and in particular Recurrent Neural Networks – RNNs) significantly enhanced natural language processing, enabling the representation of the meaning of text in a more meaningful way, valuing the context of a word in a sentence.

This is the technology that powered the virtual assistants born in the first decade of the new century, very proficient in interpreting human language, identifying a need, and performing an action to satisfy it – like answering with a pre-defined script or consuming a 3rd party service.

Farhad Rahimi

Researcher

How do large language models work?

let’s have a look at how large language models work, with a focus on OpenAI GPT (Generative Pre-trained Transformer) models.

• Tokenizer, text to numbers: Large Language Models receive a text as input and generate a text as output. However, being statistical models, they work much better with numbers than text sequences. That’s why every input to the model is processed by a tokenizer, before being used by the core model. A token is a chunk of text – consisting of a variable number of characters, so the tokenizer's main task is splitting the input into an array of tokens. Then, each token is mapped with a token index, which is the integer encoding of the original text chunk.

• Predicting output tokens: Given n tokens as input (with max n varying from one model to another), the model is able to predict one token as output. This token is then incorporated into the input of the next iteration, in an expanding window pattern, enabling a better user experience of getting one (or multiple) sentence as an answer.

• Selection process, probability distribution: The output token is chosen by the model according to its probability of occurring after the current text sequence. This is because the model predicts a probability distribution over all possible ‘next tokens’, calculated based on its training. However, not always is the token with the highest probability chosen from the resulting distribution. A degree of randomness is added to this choice, in a way that the model acts in a non-deterministic fashion - we do not get the exact same output for the same input. This degree of randomness is added to simulate the process of creative thinking and it can be tuned using a model parameter called temperature.

Farhad Rahimi

Researcher

𝗧𝗿𝗮𝗱𝗶𝘁𝗶𝗼𝗻𝗮𝗹 𝗔𝗜, 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗔𝗜, and 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜

🔹 𝗧𝗿𝗮𝗱𝗶𝘁𝗶𝗼𝗻𝗮𝗹 𝗔𝗜: 𝗧𝗵𝗲 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻

This is where AI first made its mark—solving structured, rules-driven problems.

• 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝘃𝗲 𝗔𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀: Forecast trends using historical data.

• 𝗖𝗹𝗮𝘀𝘀𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗦𝘆𝘀𝘁𝗲𝗺𝘀: Sort and categorize information automatically.

• 𝗔𝗻𝗼𝗺𝗮𝗹𝘆 𝗗𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻: Catch unusual patterns early—especially useful for fraud, risk, and system failures.

Traditional AI is reliable and rule-oriented, but limited when tasks require creativity or adaptability.

🔹 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗔𝗜: 𝗖𝗿𝗲𝗮𝘁𝗶𝗻𝗴 𝗡𝗲𝘄 𝗖𝗼𝗻𝘁𝗲𝗻𝘁

This layer goes beyond analysis—it creates.

• 𝗖𝗼𝗻𝘁𝗲𝗻𝘁 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻: Drafts, summaries, designs, and more.

• 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻: Meeting notes, email sorting, documentation.

• 𝗞𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 (𝗥𝗔𝗚): Use your internal data to power context-aware answers.

Generative AI accelerates output and enhances decision-making without fully taking over processes.

🔹 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜: 𝗧𝗵𝗲 𝗡𝗲𝘅𝘁 𝗦𝘁𝗲𝗽

This is where AI begins to act, not just analyze or generate.

• 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 & 𝗧𝗼𝗼𝗹 𝗨𝘀𝗲: Systems that trigger actions across tools and APIs.

• 𝗠𝘂𝗹𝘁𝗶-𝗔𝗴𝗲𝗻𝘁 𝗢𝗿𝗰𝗵𝗲𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻: Multiple agents working together for complex workflows.

• 𝗔𝗜 𝗣𝗿𝗼𝗱𝘂𝗰𝘁 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻: Embedding AI logic directly into products and services.

Agentic AI introduces autonomy—handling tasks with memory, reasoning, and coordination.

𝗪𝗵𝗲𝗿𝗲 𝘁𝗵𝗲 𝗥𝗲𝗮𝗹 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗩𝗮𝗹𝘂𝗲 𝗖𝗼𝗺𝗲𝘀 𝗜𝗻

• 𝗧𝗿𝗮𝗱𝗶𝘁𝗶𝗼𝗻𝗮𝗹 𝗔𝗜: Automates simple, rule-based tasks.

• 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗔𝗜: Enhances creativity, speeds up content-driven work.

• 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜: Orchestrates workflows across teams, tools, and systems.

Understanding these layers helps you decide what to adopt now and what to prepare for next.

Farhad Rahimi

Researcher

𝟴 𝗧𝘆𝗽𝗲𝘀 𝗼𝗳 𝗟𝗟𝗠𝘀 𝗣𝗼𝘄𝗲𝗿𝗶𝗻𝗴 𝗧𝗼𝗱𝗮𝘆’𝘀 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀

As AI agents become smarter and more capable, the models behind them are evolving just as quickly. We’re no longer relying on one “all-purpose” model — modern AI systems use a mix of specialized architectures built for reasoning, vision, planning, action, and deeper concept understanding.

Here’s a simple, clear rundown of the 8 major LLM types shaping the next wave of AI agents:

🔵 𝟭. 𝗚𝗣𝗧 – 𝗚𝗲𝗻𝗲𝗿𝗮𝗹 𝗣𝗿𝗲𝘁𝗿𝗮𝗶𝗻𝗲𝗱 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗲𝗿𝘀
The familiar backbone. Great at generating text, answering questions, and holding conversations thanks to large-scale pretraining.

🟢 𝟮. 𝗠𝗼𝗘 – 𝗠𝗶𝘅𝘁𝘂𝗿𝗲 𝗼𝗳 𝗘𝘅𝗽𝗲𝗿𝘁𝘀
Think of it as a team of specialists. A gating system picks the right expert for each task, making it efficient and highly scalable.

🔷 𝟯. 𝗟𝗥𝗠 – 𝗟𝗮𝗿𝗴𝗲 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗠𝗼𝗱𝗲𝗹𝘀
Built for deeper thinking. They break problems down, reason step-by-step, and validate their own answers.

🟠 𝟰. 𝗩𝗟𝗠 – 𝗩𝗶𝘀𝗶𝗼𝗻-𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹𝘀
The bridge between images and text. Essential for multimodal agents, robotics, and any AI that needs to “see” and “understand.”

🟣 𝟱. 𝗦𝗟𝗠 – 𝗦𝗺𝗮𝗹𝗹 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹𝘀
Compact, fast, and cost-effective. Perfect for edge devices, private deployments, or systems that need quick responses.

🟡 𝟲. 𝗟𝗔𝗠 – 𝗟𝗮𝗿𝗴𝗲 𝗔𝗰𝘁𝗶𝗼𝗻 𝗠𝗼𝗱𝗲𝗹𝘀
Focused on doing, not just knowing. They plan, call tools, interact with APIs, and work toward a goal.

🔵 𝟳. 𝗛𝗥𝗠 – 𝗛𝗶𝗲𝗿𝗮𝗿𝗰𝗵𝗶𝗰𝗮𝗹 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗠𝗼𝗱𝗲𝗹𝘀
Combine big-picture planning with fast local decision-making — ideal for long, multi-step tasks.

🟧 𝟴. 𝗟𝗖𝗠 – 𝗟𝗮𝗿𝗴𝗲 𝗖𝗼𝗻𝗰𝗲𝗽𝘁 𝗠𝗼𝗱𝗲𝗹𝘀
Go a layer deeper. They represent richer world concepts and generate more coherent outputs using structured decoding and diffusion-like approaches.

Farhad Rahimi

Researcher

40 Large Language Model Benchmarks and The Future of Model Evaluation

40 LLM Benchmarks