PHI 4 is a new 14 billion parameter model from Microsoft that is claimed to beat GPT-4o, Llama 3.3 and Claude 3.5 in math, and is comparable in other benchmarks. What makes this model so good however (for it's size), is its chain of thoughts
In this video we get started with PHI 4 by getting up and running locally using Ollama and testing it withon OpenWebUI. We look at its math capabilities, it's coding capabilities, it's role playing capabilities and the chain of thoughts it generates for puzzles such as sudoku or tic-tac-toe, comparing with Deep Seek R1, Claude 3.5 Sonnect, GPT-4o-mini and Llama 3.3 in various tasks. We also look at how Microsoft is priming itself for agents but isn't quite there due to lack of tool calls.
00:00 - Introduction
00:55 - Installing PHI-4 with Ollama
02:12 - Using OpenWebUI
03:06 - Math Problems
06:48 - Sudoku
11:38 - Tic-Tac-Toe
15:05 - Function Calling and Tools
17:20 - Role Playing
18:55 - Creating Stories
21:09 - Coding
24:37 - Final Thoughts
