https://mklab.gr/ Sat, 17 Oct 2026 04:23:12 +0000 MyBB https://mklab.gr/showthread.php?tid=2228 Mon, 28 Sep 2026 23:41:33 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=2228 Stanford CS329A: Self-Improving AI Agents focuses on developing artificial intelligence agents capable of continuously improving their performance through interaction with their environment. It explores techniques such as reinforcement learning (RL), test-time computational scaling, automated verification, search algorithms and feedback-based learning. 

The course also examines how large language models (LLMs) can be enhanced with memory, external tools, code execution and multimodal capabilities. Particular emphasis is placed on enabling AI systems to perform complex, multi-step reasoning, planning and autonomous tasks, rather than simply answering individual questions.

The course combines research papers, practical assignments, invited lectures and original research projects. Students investigate applications such as autonomous coding agents, scientific research assistants and robotic systems. Topics include self-improving AI, automated software engineering, long-term memory, deep research agents and the evaluation of autonomous systems. 

Students work in teams of two to four to develop original research projects, supported by API credits. Assessment consists of three homework assignments, presentations and a final research project. Overall, the course explores how AI can move beyond static models toward autonomous systems that learn, reason, use tools and improve themselves.

COURSE]]> Stanford CS329A: Self-Improving AI Agents focuses on developing artificial intelligence agents capable of continuously improving their performance through interaction with their environment. It explores techniques such as reinforcement learning (RL), test-time computational scaling, automated verification, search algorithms and feedback-based learning. 

The course also examines how large language models (LLMs) can be enhanced with memory, external tools, code execution and multimodal capabilities. Particular emphasis is placed on enabling AI systems to perform complex, multi-step reasoning, planning and autonomous tasks, rather than simply answering individual questions.

The course combines research papers, practical assignments, invited lectures and original research projects. Students investigate applications such as autonomous coding agents, scientific research assistants and robotic systems. Topics include self-improving AI, automated software engineering, long-term memory, deep research agents and the evaluation of autonomous systems. 

Students work in teams of two to four to develop original research projects, supported by API credits. Assessment consists of three homework assignments, presentations and a final research project. Overall, the course explores how AI can move beyond static models toward autonomous systems that learn, reason, use tools and improve themselves.

COURSE]]> https://mklab.gr/showthread.php?tid=2052 Sun, 20 Sep 2026 04:22:10 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=2052 This video, hosted by Andrew Ng from DeepLearningAI, provides a comprehensive prompt engineering course designed to help users transition from AI novices to power users. Throughout the course, Ng covers essential techniques for maximizing modern AI models—such as giving AI adequate time to think for complex problems, providing rich background context, understanding pre-trained knowledge limits and typo resilience, and leveraging web search and multi-step agentic workflows. 

Additionally, it demonstrates practical applications like iterative brainstorming, compiling thorough research reports, and building interactive quiz apps based on generated outputs, encouraging viewers to continually experiment with new models and tackle challenging tasks.


COURSE]]> This video, hosted by Andrew Ng from DeepLearningAI, provides a comprehensive prompt engineering course designed to help users transition from AI novices to power users. Throughout the course, Ng covers essential techniques for maximizing modern AI models—such as giving AI adequate time to think for complex problems, providing rich background context, understanding pre-trained knowledge limits and typo resilience, and leveraging web search and multi-step agentic workflows. 

Additionally, it demonstrates practical applications like iterative brainstorming, compiling thorough research reports, and building interactive quiz apps based on generated outputs, encouraging viewers to continually experiment with new models and tackle challenging tasks.


COURSE]]> https://mklab.gr/showthread.php?tid=2016 Wed, 16 Sep 2026 01:30:35 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=2016 freeCodeCamp 65-hour Generative AI Bootcamp is a comprehensive, project-oriented course designed to take learners from foundational GenAI concepts to practical application development. It covers Python and Jupyter Notebooks, data libraries, prompt engineering, model fine-tuning, GenAI architecture, APIs, frontend/backend development, security, and AI agents. Rather than focusing only on theory, the course has learners build real applications, including AI-powered language-learning tools, while introducing technologies such as Ollama, AWS Lambda, DeepSeek, Gemini, and OPEA. 

The later sections become increasingly hands-on, covering multimodal AI, structured JSON outputs, containers, microservices, text-to-speech, transcription with WhisperX, AI agents, and image-generation applications. The bootcamp also includes guest instructors and discussions about responsible AI, security, and how companies deploy AI systems. Overall, it is structured as a practical introduction-to-intermediate pathway, with a strong emphasis on building a portfolio of working GenAI projects rather than simply learning terminology. 

COURSE]]> freeCodeCamp 65-hour Generative AI Bootcamp is a comprehensive, project-oriented course designed to take learners from foundational GenAI concepts to practical application development. It covers Python and Jupyter Notebooks, data libraries, prompt engineering, model fine-tuning, GenAI architecture, APIs, frontend/backend development, security, and AI agents. Rather than focusing only on theory, the course has learners build real applications, including AI-powered language-learning tools, while introducing technologies such as Ollama, AWS Lambda, DeepSeek, Gemini, and OPEA. 

The later sections become increasingly hands-on, covering multimodal AI, structured JSON outputs, containers, microservices, text-to-speech, transcription with WhisperX, AI agents, and image-generation applications. The bootcamp also includes guest instructors and discussions about responsible AI, security, and how companies deploy AI systems. Overall, it is structured as a practical introduction-to-intermediate pathway, with a strong emphasis on building a portfolio of working GenAI projects rather than simply learning terminology. 

COURSE]]> https://mklab.gr/showthread.php?tid=2013 Tue, 15 Sep 2026 23:03:04 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=2013 CIS 6280: World Models course, taught by Jiatao Gu in Fall 2026, explores how learned models of environments can support perception, prediction, planning, reasoning, and decision-making. It begins with the foundations of world modeling: observations, latent states, transitions, memory, actions, simulation, and probabilistic formulations. 

Students then study environments and rollouts, state-space models, representation learning, and modern generative-model foundations, including VAEs, GANs, autoregressive models, diffusion models, flow matching, and normalizing flows.

The second half moves from general generative modeling toward applications of world models. Topics include sequential world models and model-based reinforcement learning, video world models, 3D/spatial representations, learned physical dynamics, robotics and sim-to-real transfer, vision-language-action systems, LLMs as world models, reasoning models, and world models for digital agents such as games and software environments. 

The course concludes by examining how world models should be evaluated—including utility, robustness, calibration, out-of-distribution behavior, intervention, drift, and failure—followed by student projects and a final exam. Overall, the syllabus presents world models as a unifying framework connecting generative AI, reinforcement learning, multimodal agents, reasoning, and robotics. 

COURSE]]> CIS 6280: World Models course, taught by Jiatao Gu in Fall 2026, explores how learned models of environments can support perception, prediction, planning, reasoning, and decision-making. It begins with the foundations of world modeling: observations, latent states, transitions, memory, actions, simulation, and probabilistic formulations. 

Students then study environments and rollouts, state-space models, representation learning, and modern generative-model foundations, including VAEs, GANs, autoregressive models, diffusion models, flow matching, and normalizing flows.

The second half moves from general generative modeling toward applications of world models. Topics include sequential world models and model-based reinforcement learning, video world models, 3D/spatial representations, learned physical dynamics, robotics and sim-to-real transfer, vision-language-action systems, LLMs as world models, reasoning models, and world models for digital agents such as games and software environments. 

The course concludes by examining how world models should be evaluated—including utility, robustness, calibration, out-of-distribution behavior, intervention, drift, and failure—followed by student projects and a final exam. Overall, the syllabus presents world models as a unifying framework connecting generative AI, reinforcement learning, multimodal agents, reasoning, and robotics. 

COURSE]]> https://mklab.gr/showthread.php?tid=2012 Tue, 15 Sep 2026 22:55:14 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=2012 This YouTube playlist is an educational lecture series titled "Mathematical Foundations of Generative AI," featuring content originally offered by NPTEL from the Indian Institute of Science, Bengaluru. Designed for students, researchers, and tech aspirants, the curriculum offers a deep dive into the theoretical and mathematical principles that power modern generative artificial intelligence models.

Spanning 18 structured lectures and tutorials, the series covers advanced topics such as Variational Divergence Minimization, $f$-Divergence, and the mechanics behind various Generative Adversarial Networks including Vanilla GANs, DCGANs, and Conditional GANs. It provides a comprehensive academic roadmap for learners looking to master the rigorous mathematical framework underlying generative modeling without getting lost in standard internet distractions.

PLAYLIST

COURSE LINK]]> This YouTube playlist is an educational lecture series titled "Mathematical Foundations of Generative AI," featuring content originally offered by NPTEL from the Indian Institute of Science, Bengaluru. Designed for students, researchers, and tech aspirants, the curriculum offers a deep dive into the theoretical and mathematical principles that power modern generative artificial intelligence models.

Spanning 18 structured lectures and tutorials, the series covers advanced topics such as Variational Divergence Minimization, $f$-Divergence, and the mechanics behind various Generative Adversarial Networks including Vanilla GANs, DCGANs, and Conditional GANs. It provides a comprehensive academic roadmap for learners looking to master the rigorous mathematical framework underlying generative modeling without getting lost in standard internet distractions.

PLAYLIST

COURSE LINK]]> https://mklab.gr/showthread.php?tid=2010 Tue, 15 Sep 2026 22:44:33 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=2010 Modeling: MultiModal AI (MMAI) course for Spring 2026 focuses on how AI can work with multiple types of data simultaneously rather than treating language, images, audio, sensors, medical information, music, and other modalities separately. 

The course covers core techniques for representing and fusing heterogeneous data, aligning different views, performing multimodal reasoning, generating multimodal content, transferring knowledge between high- and low-resource domains, and developing safe, ethical, human-aligned AI systems. Its broader goal is to give students both technical knowledge of recent AI developments and the intuition needed to apply AI to unfamiliar combinations of data.

The course is strongly hands-on and research-oriented. Students attend two lectures per week and complete readings, homework, and intermediate practical assignments in which they apply multimodal AI to data and problems of their choice. The course culminates in a team-based, instructor-approved research project investigating a new data modality or a multimodal AI problem. In this sense, the course is not simply an introduction to existing AI models: it is designed to teach students how to think like AI researchers, formulate new problems, experiment with multimodal systems, and potentially create novel AI methods or applications.

COURSE LINK]]> Modeling: MultiModal AI (MMAI) course for Spring 2026 focuses on how AI can work with multiple types of data simultaneously rather than treating language, images, audio, sensors, medical information, music, and other modalities separately. 

The course covers core techniques for representing and fusing heterogeneous data, aligning different views, performing multimodal reasoning, generating multimodal content, transferring knowledge between high- and low-resource domains, and developing safe, ethical, human-aligned AI systems. Its broader goal is to give students both technical knowledge of recent AI developments and the intuition needed to apply AI to unfamiliar combinations of data.

The course is strongly hands-on and research-oriented. Students attend two lectures per week and complete readings, homework, and intermediate practical assignments in which they apply multimodal AI to data and problems of their choice. The course culminates in a team-based, instructor-approved research project investigating a new data modality or a multimodal AI problem. In this sense, the course is not simply an introduction to existing AI models: it is designed to teach students how to think like AI researchers, formulate new problems, experiment with multimodal systems, and potentially create novel AI methods or applications.

COURSE LINK]]> https://mklab.gr/showthread.php?tid=2007 Tue, 15 Sep 2026 21:18:53 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=2007 2012 Coursera course, Neural Networks for Machine Learning is a comprehensive introduction to the ideas that underpin modern deep learning. It begins with the motivation for machine learning and basic neuron models, then develops perceptrons, linear and logistic neurons, gradient-based learning, and the backpropagation algorithm. 

The course subsequently moves into language modeling and word prediction, convolutional neural networks (CNNs) for image recognition, mini-batch gradient descent and optimization methods such as momentum and RMSProp, and recurrent neural networks (RNNs) and LSTMs for sequential data. 

The second half focuses increasingly on the foundations of deep learning and probabilistic models. Hinton covers regularization and generalization, Bayesian approaches, mixtures of experts and dropout, followed by Hopfield networks, Boltzmann machines and restricted Boltzmann machines (RBMs). 

These lead into belief networks, the wake-sleep algorithm, deep belief networks, layer-wise pretraining and fine-tuning, and finally autoencoders, deep autoencoders, semantic hashing, multimodal image-caption models, hierarchical representations, and Bayesian hyperparameter optimization. Although the lectures date from 2012, they provide an unusually valuable historical and conceptual foundation for understanding how neural-network research evolved into modern deep learning. 

LECTURES]]> 2012 Coursera course, Neural Networks for Machine Learning is a comprehensive introduction to the ideas that underpin modern deep learning. It begins with the motivation for machine learning and basic neuron models, then develops perceptrons, linear and logistic neurons, gradient-based learning, and the backpropagation algorithm. 

The course subsequently moves into language modeling and word prediction, convolutional neural networks (CNNs) for image recognition, mini-batch gradient descent and optimization methods such as momentum and RMSProp, and recurrent neural networks (RNNs) and LSTMs for sequential data. 

The second half focuses increasingly on the foundations of deep learning and probabilistic models. Hinton covers regularization and generalization, Bayesian approaches, mixtures of experts and dropout, followed by Hopfield networks, Boltzmann machines and restricted Boltzmann machines (RBMs). 

These lead into belief networks, the wake-sleep algorithm, deep belief networks, layer-wise pretraining and fine-tuning, and finally autoencoders, deep autoencoders, semantic hashing, multimodal image-caption models, hierarchical representations, and Bayesian hyperparameter optimization. Although the lectures date from 2012, they provide an unusually valuable historical and conceptual foundation for understanding how neural-network research evolved into modern deep learning. 

LECTURES]]> https://mklab.gr/showthread.php?tid=2005 Tue, 15 Sep 2026 21:09:53 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=2005 CS329Z: Engineering AI Agents is a Fall 2026 course focused on the practical engineering of agentic AI systems—moving beyond standalone LLMs toward systems that combine language models with retrieval, tools, memory, optimizers, and multiple interacting agents. The course covers the full development process: deciding when agents are appropriate, decomposing tasks, selecting models and components, engineering context, building RAG systems and tool-using agents, designing agent workflows and multi-agent architectures, optimizing prompts/models, collecting data, and creating reliable evaluations. It also addresses safety, privacy, prompt injection, sandboxing, coding agents, proactive agents, multimodal/web agents, and the challenges of deploying long-running systems. 

A particularly strong aspect is its hands-on engineering emphasis. Students first build fundamental components such as RAG pipelines, tool-use systems, memory, and agent loops from scratch, before using frameworks such as DSPy, LangGraph, and LlamaIndex. They then build an agentic harness, design a comprehensive evaluation system for an agent, analyze recent research, and complete a quarter-long project involving a working agent.

 The course therefore treats agents not simply as prompting exercises but as systems that must be designed, optimized, evaluated, and made reliable. Its central message is that successful AI applications increasingly depend on the engineering of the surrounding system—not merely on choosing a more powerful language model.

COURSE]]> CS329Z: Engineering AI Agents is a Fall 2026 course focused on the practical engineering of agentic AI systems—moving beyond standalone LLMs toward systems that combine language models with retrieval, tools, memory, optimizers, and multiple interacting agents. The course covers the full development process: deciding when agents are appropriate, decomposing tasks, selecting models and components, engineering context, building RAG systems and tool-using agents, designing agent workflows and multi-agent architectures, optimizing prompts/models, collecting data, and creating reliable evaluations. It also addresses safety, privacy, prompt injection, sandboxing, coding agents, proactive agents, multimodal/web agents, and the challenges of deploying long-running systems. 

A particularly strong aspect is its hands-on engineering emphasis. Students first build fundamental components such as RAG pipelines, tool-use systems, memory, and agent loops from scratch, before using frameworks such as DSPy, LangGraph, and LlamaIndex. They then build an agentic harness, design a comprehensive evaluation system for an agent, analyze recent research, and complete a quarter-long project involving a working agent.

 The course therefore treats agents not simply as prompting exercises but as systems that must be designed, optimized, evaluated, and made reliable. Its central message is that successful AI applications increasingly depend on the engineering of the surrounding system—not merely on choosing a more powerful language model.

COURSE]]> https://mklab.gr/showthread.php?tid=2004 Tue, 15 Sep 2026 20:57:43 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=2004 “Artificial Intelligence for All” guide is a practical introduction to using AI in everyday life and at work, designed specifically for people without technical expertise. It treats effective communication with AI systems as an emerging form of digital literacy and aims to give citizens practical skills while helping them understand what AI can and cannot do. The guide is organized around several core areas: understanding AI and large language models, learning how to write effective prompts, exploring practical applications, and using AI responsibly. (

A major emphasis is therefore placed on practical prompting and responsible AI use. The guide provides examples across eight application categories, with more than 25 practical examples, while also addressing AI’s limitations, risks, ethical considerations, and best practices. It additionally includes a glossary and other useful resources, and is freely accessible, reflecting the government's goal of making AI knowledge broadly available rather than limiting it to technical specialists.

COURSE]]> “Artificial Intelligence for All” guide is a practical introduction to using AI in everyday life and at work, designed specifically for people without technical expertise. It treats effective communication with AI systems as an emerging form of digital literacy and aims to give citizens practical skills while helping them understand what AI can and cannot do. The guide is organized around several core areas: understanding AI and large language models, learning how to write effective prompts, exploring practical applications, and using AI responsibly. (

A major emphasis is therefore placed on practical prompting and responsible AI use. The guide provides examples across eight application categories, with more than 25 practical examples, while also addressing AI’s limitations, risks, ethical considerations, and best practices. It additionally includes a glossary and other useful resources, and is freely accessible, reflecting the government's goal of making AI knowledge broadly available rather than limiting it to technical specialists.

COURSE]]> https://mklab.gr/showthread.php?tid=1974 Tue, 15 Sep 2026 17:55:17 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=1974 COS 324: Introduction to Machine Learning notes from Princeton University provide a mathematically grounded introduction to the fundamental ideas of machine learning. The course begins with supervised learning, covering linear regression, statistical learning theory, gradient descent, and linear classification, before moving to unsupervised learning, including clustering, dimensionality reduction, language models based on n-grams, and matrix factorization/recommender systems. The notes were prepared by Sanjeev Arora, Danqi Chen, and Princeton students, and were last updated in September 2024. 

The second half introduces deep learning and reinforcement learning, including neural networks, backpropagation, convolutional neural networks, Markov decision processes, and learning in unknown environments. It also includes advanced topics such as machine-learning ethics and NLP, followed by dedicated mathematical foundations in probability and statistics, calculus, and linear algebra. Overall, the material is particularly useful for someone seeking a rigorous university-level foundation: it connects the mathematical principles behind machine-learning algorithms with modern topics such as deep learning and reinforcement learning, rather than treating machine learning purely as a collection of programming techniques. 

Princeton COS 324 course notes]]> COS 324: Introduction to Machine Learning notes from Princeton University provide a mathematically grounded introduction to the fundamental ideas of machine learning. The course begins with supervised learning, covering linear regression, statistical learning theory, gradient descent, and linear classification, before moving to unsupervised learning, including clustering, dimensionality reduction, language models based on n-grams, and matrix factorization/recommender systems. The notes were prepared by Sanjeev Arora, Danqi Chen, and Princeton students, and were last updated in September 2024. 

The second half introduces deep learning and reinforcement learning, including neural networks, backpropagation, convolutional neural networks, Markov decision processes, and learning in unknown environments. It also includes advanced topics such as machine-learning ethics and NLP, followed by dedicated mathematical foundations in probability and statistics, calculus, and linear algebra. Overall, the material is particularly useful for someone seeking a rigorous university-level foundation: it connects the mathematical principles behind machine-learning algorithms with modern topics such as deep learning and reinforcement learning, rather than treating machine learning purely as a collection of programming techniques. 

Princeton COS 324 course notes]]> https://mklab.gr/showthread.php?tid=1890 Tue, 08 Sep 2026 00:20:01 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=1890 Source: LangChain Academy
Course:Quickstart: LangGraph Essentials – Python

This short course is a practical introduction to LangGraph, LangChain’s low-level framework for building stateful and controllable AI-agent workflows in Python. Rather than treating an agent as a simple linear chain of prompts, LangGraph models an application as a graph whose components maintain and modify shared state. The course introduces the central primitives: nodes, where computation or LLM/tool operations occur; edges, which determine how execution moves between nodes; and conditional edges, which allow the workflow to choose different paths depending on the current state or an agent’s decision. 

The course then moves to two especially important capabilities for agentic systems: memory and human-in-the-loop control. LangGraph can persist graph state through checkpoints, allowing an agent to retain conversational or workflow context and resume execution later. The mechanism allows execution to pause so a person can review, approve, modify, or provide additional information before the workflow continues. These features make LangGraph particularly useful for applications that are more complex than ordinary chatbot loops—for example, research agents, approval workflows, multi-stage automation, and agents performing potentially consequential actions. 

The final lesson brings these elements together into an application, showing the underlying architecture of a LangGraph program: define the application state, implement nodes that transform that state, connect them with normal or conditional edges, add persistence where needed, and compile the resulting graph into an executable workflow. LangGraph is therefore best understood not simply as another LLM library, but as an orchestration/runtime layer for sophisticated agents, particularly when developers need precise control over execution, persistent state, branching logic, interruptions, and durable workflows. LangChain itself recommends LangGraph for advanced cases combining deterministic application logic with agentic behavior. 

Course structure

1. Nodes — individual units of computation.
   
2. Edges — transitions between nodes.
   
3. Conditional Edges — dynamic branching of execution.
   
4. Memory — persistence of state and conversation context.
   
5. Interrupt / Human-in-the-Loop — pausing and resuming workflows with human intervention.
   
6. Application — combining the concepts into a working LangGraph program.
   

COURSE]]> Source: LangChain Academy
Course:Quickstart: LangGraph Essentials – Python

This short course is a practical introduction to LangGraph, LangChain’s low-level framework for building stateful and controllable AI-agent workflows in Python. Rather than treating an agent as a simple linear chain of prompts, LangGraph models an application as a graph whose components maintain and modify shared state. The course introduces the central primitives: nodes, where computation or LLM/tool operations occur; edges, which determine how execution moves between nodes; and conditional edges, which allow the workflow to choose different paths depending on the current state or an agent’s decision. 

The course then moves to two especially important capabilities for agentic systems: memory and human-in-the-loop control. LangGraph can persist graph state through checkpoints, allowing an agent to retain conversational or workflow context and resume execution later. The mechanism allows execution to pause so a person can review, approve, modify, or provide additional information before the workflow continues. These features make LangGraph particularly useful for applications that are more complex than ordinary chatbot loops—for example, research agents, approval workflows, multi-stage automation, and agents performing potentially consequential actions. 

The final lesson brings these elements together into an application, showing the underlying architecture of a LangGraph program: define the application state, implement nodes that transform that state, connect them with normal or conditional edges, add persistence where needed, and compile the resulting graph into an executable workflow. LangGraph is therefore best understood not simply as another LLM library, but as an orchestration/runtime layer for sophisticated agents, particularly when developers need precise control over execution, persistent state, branching logic, interruptions, and durable workflows. LangChain itself recommends LangGraph for advanced cases combining deterministic application logic with agentic behavior. 

Course structure

1. Nodes — individual units of computation.
   
2. Edges — transitions between nodes.
   
3. Conditional Edges — dynamic branching of execution.
   
4. Memory — persistence of state and conversation context.
   
5. Interrupt / Human-in-the-Loop — pausing and resuming workflows with human intervention.
   
6. Application — combining the concepts into a working LangGraph program.
   

COURSE]]> https://mklab.gr/showthread.php?tid=1875 Mon, 07 Sep 2026 20:21:51 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=1875 AI coding agent from scratch in Python using the free Gemini API. Rather than creating a simple chatbot, the project focuses on the core mechanics of modern agentic coding systems: implementing an agent loop, using tool/function calling, and allowing the model to read files, write or modify code, and execute programs inside a project. 

The finished agent is then tested on a practical task—identifying and fixing bugs in an existing program—giving learners a concrete understanding of how tools such as autonomous coding assistants operate behind the scenes. The roughly two-hour course is aimed at developers who want to strengthen both their Python skills and their understanding of agentic AI by building a functional coding agent themselves. 

COURSE]]> AI coding agent from scratch in Python using the free Gemini API. Rather than creating a simple chatbot, the project focuses on the core mechanics of modern agentic coding systems: implementing an agent loop, using tool/function calling, and allowing the model to read files, write or modify code, and execute programs inside a project. 

The finished agent is then tested on a practical task—identifying and fixing bugs in an existing program—giving learners a concrete understanding of how tools such as autonomous coding assistants operate behind the scenes. The roughly two-hour course is aimed at developers who want to strengthen both their Python skills and their understanding of agentic AI by building a functional coding agent themselves. 

COURSE]]> https://mklab.gr/showthread.php?tid=1874 Mon, 07 Sep 2026 20:18:43 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=1874 six-hour hands-on course by Vivek Kalyanarangan that teaches how to build a Large Language Model from scratch using pure PyTorch, combining the underlying theory with practical implementation. It begins with the core Transformer architecture and training a small language model, then adds modern techniques such as RMSNorm, Rotary Positional Embeddings (RoPE), KV caching, mixed-precision training, and Mixture-of-Experts (MoE). 
The course then moves into post-training and alignment, covering Supervised Fine-Tuning (SFT), reward modeling, and Reinforcement Learning from Human Feedback (RLHF) using PPO, showing how a base model can be shaped into a more useful and safer assistant. Overall, it is designed as an end-to-end introduction to how modern LLMs are constructed, optimized, scaled, fine-tuned, and aligned, with the complete code available for experimentation.


COURSE]]> six-hour hands-on course by Vivek Kalyanarangan that teaches how to build a Large Language Model from scratch using pure PyTorch, combining the underlying theory with practical implementation. It begins with the core Transformer architecture and training a small language model, then adds modern techniques such as RMSNorm, Rotary Positional Embeddings (RoPE), KV caching, mixed-precision training, and Mixture-of-Experts (MoE). 
The course then moves into post-training and alignment, covering Supervised Fine-Tuning (SFT), reward modeling, and Reinforcement Learning from Human Feedback (RLHF) using PPO, showing how a base model can be shaped into a more useful and safer assistant. Overall, it is designed as an end-to-end introduction to how modern LLMs are constructed, optimized, scaled, fine-tuned, and aligned, with the complete code available for experimentation.


COURSE]]> https://mklab.gr/showthread.php?tid=1872 Mon, 07 Sep 2026 20:13:38 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=1872 fine-tuning large language models (LLMs), explaining how fine-tuning adapts a pretrained model to specialized tasks or domains by training it further on targeted data. It distinguishes fine-tuning from pre-training and prompt engineering, covers methods such as supervised fine-tuning and reinforcement learning from human feedback (RLHF), and highlights parameter-efficient fine-tuning (PEFT) techniques that reduce computational requirements. 
A major focus is QLoRA, which combines quantization with low-rank adaptation to make it possible to fine-tune very large models, such as Llama 70B, on comparatively modest hardware. The course also walks through the practical fine-tuning workflow, benefits and limitations, and real-world case studies, aiming to give learners both the conceptual foundations and hands-on skills needed to customize LLMs for applications such as specialized chatbots or domain-specific AI assistants. 

COURSE]]> fine-tuning large language models (LLMs), explaining how fine-tuning adapts a pretrained model to specialized tasks or domains by training it further on targeted data. It distinguishes fine-tuning from pre-training and prompt engineering, covers methods such as supervised fine-tuning and reinforcement learning from human feedback (RLHF), and highlights parameter-efficient fine-tuning (PEFT) techniques that reduce computational requirements. 
A major focus is QLoRA, which combines quantization with low-rank adaptation to make it possible to fine-tune very large models, such as Llama 70B, on comparatively modest hardware. The course also walks through the practical fine-tuning workflow, benefits and limitations, and real-world case studies, aiming to give learners both the conceptual foundations and hands-on skills needed to customize LLMs for applications such as specialized chatbots or domain-specific AI assistants. 

COURSE]]> https://mklab.gr/showthread.php?tid=1562 Tue, 11 Aug 2026 11:10:33 +0300 mklabgr]]> https://mklab.gr/showthread.php?tid=1562 CSE 573 Introduction to AI

Summary

The University of Washington’s CSE 573: Introduction to Artificial Intelligence is a comprehensive course covering the core ideas and algorithms behind modern AI. The curriculum progresses from intelligent agents and problem-solving/search algorithms to informed and adversarial search, ExpectiMax, Markov Decision Processes, reinforcement learning, uncertainty, Hidden Markov Models, Bayesian networks and probabilistic inference, and machine learning. 

Students apply these concepts through programming projects involving search, multi-agent systems, reinforcement learning, and inference/filtering, alongside a written assignment and a final research or practical AI project. The course recommends Russell and Norvig’s Artificial Intelligence: A Modern Approach as its main textbook and combines theoretical foundations with substantial hands-on programming.


COURSE]]> CSE 573 Introduction to AI

Summary

The University of Washington’s CSE 573: Introduction to Artificial Intelligence is a comprehensive course covering the core ideas and algorithms behind modern AI. The curriculum progresses from intelligent agents and problem-solving/search algorithms to informed and adversarial search, ExpectiMax, Markov Decision Processes, reinforcement learning, uncertainty, Hidden Markov Models, Bayesian networks and probabilistic inference, and machine learning. 

Students apply these concepts through programming projects involving search, multi-agent systems, reinforcement learning, and inference/filtering, alongside a written assignment and a final research or practical AI project. The course recommends Russell and Norvig’s Artificial Intelligence: A Modern Approach as its main textbook and combines theoretical foundations with substantial hands-on programming.


COURSE]]>