Скачать или смотреть Ten Tests

#GPT is showing us is that there's an additional kind of regularity to language which focuses on the meaning of the language beyond just the text generation concept.

One example of that its that is seems to have used language to uncover logic. 

Aristotle uncovered logic by realizing there's a structure to sentences. #ChatGPT is operating at the Aristotelian level. It's essentially dealing with templates of sentences.

#CodeInterpreter is not just an input and output to the GPT-4 model, but it also gives you access to a Python Execution Environment with access to a number of Python Libraries and can perform complex computations.

But adding Code to GPT also has a powerful side effect. Combining Code logic and GPT could potentially improve the models' abilities to reason and generate text in a more logically structured and coherent manner.

This video performs Ten Tests on #ChatGPT Code Interpreter alpha using Truth Tables to test the model's Boolean logic capabilities.

LEARN MORE IN THIS ARTICLE:   / large-language-models-beyond-text-generati...  

00:00-02:27 Introduction
2:28-3:37 Truth Tables Experiment
3:38-4:14 Model Used by Code Interpreter?
4:15-5:10 Truth Tables
5:11-6:03 Test One - Not P or Q
6:04-6:36 Test Two - Not (P and not Q)
6:37-6:59 Test Three - (P implies Q) or (not P implies R)
7:00-7:40 Test Four - P or Q or R
7:41-8:16 Test Five - (P and Q) implies R
8:17-8:51 Test Six - Complicated Truth Table
8:52-10:04- Remaining Tests
10:05-10:22 Conclusion