Скачать или смотреть Can In-Context Learning (ICL) Replace Fine-tuning LLMs? Real Evidence From GPT-4 🤯

#ai #gpt4o #llm
🔴 This AI Agent Skill Pays More Than Traditional Software Dev in 2025 (Here's Why) 👉    • 🔴 This AI Agent Skill Pays More Than Tradi...  

This video:
When GPT-4o or omni was released, there was so much excitement about the super cool voice and visual recognition and generation that we overlooked an important aspect of it that I will be discussing today, based on this paper by several Stanford University researchers, including Andrew NG:
Jiang et al. (2024). Many-Shot In-Context Learning in Multimodal Foundation Models

Related videos:
What Language Model To Choose For Your Project? LLM Evaluation
   • What Language Model To Choose For Your Pro...  

💯 FREE Local Set-up For AI Agents With CrewAI And Ollama | Easy Tutorial
   • 💯 FREE Local LLM - AI Agents With CrewAI A...  

   / @analyticscamp  

Chapters and Key Concepts:
00:00 Introducing GPT-4o
00:32 Live Translation
00:41 Real-time Recognition of Objects and Animals
00:50 Real-time Recognition of Historical Sites
01:07 Two GPT-4os Role-switching
01:27 Issues with Current GPT-4o Model and Fine-tuned LLMs
01:56 Many-shot In-Context-Learning (ICL) in Multimodal Foundation Models
02:30 Explaining ICL
02:48 Zero-shot, Few-shot, and Many-shot ICL with Image
03:27 Issues with Few-shot ICL and Context Window
04:21 Many-shot ICL Benchmarking with Image Datasets
04:26 Results of GPT-4o and Gemini 1.5 Pro on ICL Research
05:01 Can We Replace Fine-tuning LLMs with ICL?
05:24 ICL Prompt Formats
05:34 Advantages of Using ICL
05:58 Taxonomy of In-Context-Learning Studies and Methods
06:18 Scoring Function for ICL
07:00 Conclusions and Questions