Can AI be truly Explainable? Explainable AI (XAI) and Large Language Models (LLM) | Ai Heroes 2024

🎟️ the 2025 edition on 27 May: https://aiheroes.it/2025/tickets/

⏩ Chapter:
00:00 Intro
04:43 Evidence of emerging properties
07:08 (XAI) eXplainable AI
09:49 Why XAI?
11:58 XAI taxonomy
13:07 Agnostic Methods
14:25 Global Surrogate
15:28 SHAP
16:11 Occlusions as agnostic method
17:14 Grad-Cam as a model specific method
18:29 XAI Summary
20:41 Large Language Model
21:45 Attention Is All You Need
26:14 Temperature
27:18 XAI and LLM
29:35 Visualizing the Attention mechanism
31:42 Transformer Explainer - LLM visualization
33:32 Explainability in Multimodal Large Language Models
35:33 GSM-Symbolic benchmark
37:37 ChatGPT is bullshit
40:26 Conclusions

In this talk, we will explore the field of Explainable AI (XAI), with a particular focus on techniques used to gain insights into Large Language Models (LLMs). We will show how XAI can be applied to LLMs, specifically analyzing attention maps and the Transformer architecture. Additionally, we will walk through the process of building an LLM from scratch, using visualization tools to delve into its inner workings. The talk will address key questions surrounding the explainability of LLMs, such as: How can we interpret these vast models? Do LLMs primarily memorize or generalize? What are the emergent capabilities of LLMs, and how do they arise?
🎤 LEONIDA GIANFAGNA, CTO of Cyber Guru
🎤 ENRICO ZIMUEL, Tech Lead and Principal Software Engineer at Elastic

📍AI Heroes is the Italian Artificial Intelligence Conference for Developers, Engineers, Data Scientists & Product Managers.
- The New Workflow to Shape the Future of Business
- AI in the Next Generation Workforce
- Key Technologies, Applications and Trends

ℹ️ inform: https://aiheroes.it/
✅ Tweet: https://twitter.com/AIHeroes_it
✅ Connect: https://www.linkedin.com/showcase/aiheroes-it/
✅ Read: https://t.me/aiheroes

#AI #MachineLearning #XAI #ExplainableAI #LargeLanguageModels #DeepLearning #ArtificialIntelligence #AIExplained #TechTalk #AIHeroes

Видео Can AI be truly Explainable? Explainable AI (XAI) and Large Language Models (LLM) | Ai Heroes 2024 канала AI Heroes