Table-GPT by Microsoft: Empower LLMs To Understand Tables

In this video we dive into Table-GPT, a large language model (LLM) that can understand tables. Table-GPT was presented in a new research paper from Microsoft, titled "Table-GPT: Table-tuned GPT for Diverse Table Tasks".
We first review the capabilities of current LLMs, ChatGPT in particular, to understand tables, which shows a big room for improvement.
Then, we introduce table-tuning, a concept introduced in the paper to improve the performance of LLMs on tables related tasks. This method was inspired from instruction-tuning, and so we first to a quick recap of large language models training process.
Afterwards, we dive into the creation of the dataset used for table-tuning, which is created using a method the researchers call synthesis-then-augment, which include automatic generation of the dataset from real tables, using synthesis and augmentation.
Finally, we review the results of Table-GPT comparing to GPT 3.5 and ChatGPT.
Paper page - arxiv.org/abs/2310.09263
Read our post - aipapersacademy.com/table-gpt/
👍 Please like & subscribe if you enjoy this content
-----------------------------------------------------------------------------------------------
Support us - paypal.me/aipapersacademy
We use VideoScribe to edit our videos - tidd.ly/44TZEiX (affiliate)
We use ChatPDF to analyze research papers - www.chatpdf.com/?via=ai-papers (affiliate)
-----------------------------------------------------------------------------------------------
Chapters:
0:00 Introduction
0:54 Can Current LLMs Understand Tables?
3:02 Table-Tuning
5:00 Synthesis-then-Augment
8:10 Results

[@sek10ng]

Now knowing too much about AI, whenever I saw a video with music I thought the music must be generated by MusicGen or something similar haha

[@sanesanyo]

Is there benchmarking data available for larger LLMs like GPT4-Turbo or Claude-3-Opus?

[@SuperCombatarms]

Is there any code associated with this study?