Agentic AI L2 A Deep Dive into LLM

Agentic AI L2 A Deep Dive into LLM

by Artificial Intelligence & Data Science شرح بالعربي · View the channel on PSI

Formal & Physical Sciences AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 Artificial Intelligence & Data Science شرح بالعربي 👥 12K 📅 October 2, 2026 ⏱ 89 min 👁 11 📄 tutorial 🧭 2026-10-02
Available in: English (current) Français

Keywords

Large Language Modeltransformerself-supervised learningneural networkembedding

Summary

This video is a lecture (in Arabic) introducing Large Language Models (LLMs) for beginners. The instructor begins by contextualizing the course, explaining the relationship between AI, large language models, and agentic AI. He then describes an LLM as a large parameterized function that maps text input to text output, trained to predict the next word. The training process is broken down into self-supervised pre-training, supervised fine-tuning, and reinforcement learning from human feedback. The core architecture, the Transformer, is explained through the attention mechanism, which allows the model to weigh token relationships. The video covers essential components: tokenization, indexation, embeddings, positional encoding, and the query/key/value concept. It emphasizes how embeddings capture semantic meaning, and how attention resolves issues of long-range dependencies and parallelization. The pedagogical style uses analogies and simple examples, but no external sources or citations are provided.

138 words

Critical Evaluation

Value of the Information & Strength of the Argument

The information value is moderate; it serves as a foundational tutorial for understanding LLMs. The argumentation is clear and highly didactic, using examples like ‘king - man + woman = queen’ to explain semantic vector spaces. However, the reasoning is often verbose and occasionally digresses, and some technical details are approximated (e.g., the embedding dimension of GPT-2). The instructor builds a logical progression from basic functions to complex attention mechanisms, but lacks precise mathematical rigor.

Scientific Rigor, Source Quality, Title Accuracy

No sources are cited in the video or description. The scientific rigor is limited by the absence of references and a few inaccuracies, such as a slightly wrong description of the GPT-2 token count (actually 50257, which is correctly mentioned) and a vague number for GPT-2’s embedding size (768, which is correct). The title ‘Agentic AI L2’ is somewhat misleading because the content focuses on LLM fundamentals rather than agentic AI specifics. The video’s originality lies in its accessible approach for Arabic-speaking learners, but it adds little for an already-informed audience.

181 words

Title / Content Match

The title mentions 'Agentic AI L2' but the content is a general introduction to LLMs, not specifically focused on agentic AI. Partially adequate.

Quality & Reliability

6/10

The content provides a correct but simplified introduction to LLMs, lacking explicit references and containing some imprecisions (e.g., the exact number of GPT-2 dimensions), yet the core concepts are accurately presented.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Contribution & Novelties

The video serves as a didactic bridge for Arabic speakers, translating complex NLP concepts into accessible language with relatable examples. It brings no new research but offers a pedagogical synthesis of known material.\n\nPour aller plus loin :\n- Attention Is All You Need — The seminal paper introducing the Transformer architecture, directly referenced in the video.\n- Transformer (machine learning model) — Detailed background on the architecture, including attention and positional encoding.\n- Word embedding — Explores the vector representations that the video covers, including the king-man analogy.\n- Self-supervised learning — Provides a broader context for the training paradigm discussed in the video.

100 words

Radar Profile

The radar profile shows a relatively balanced shape with a slight peak in quantity of information, indicating a dense tutorial. The quality and reliability scores are moderate, reflecting the lack of citations and some simplifications, while the technical level is sufficient for a beginners' course.

Reliability 6/10