Is AI a Bubble?

Can AI fix real-life car problems? We test if LLMs can help repair a 1972 VW Beetle using AI, manuals, and Google data. Are the results practical?

Written by — Mika Heino, Data Architect

AI (or LLMs, to be more specific) has been part of our lives for the past two years. We've seen a multitude of examples where ChatGPT or similar tools have been used to ease the lives of white-collar workers in their daily online activities. Not all problems can be fixed by simple programming code, though.

In this episode, we will investigate whether AI or LLMs can help us (in this case, me and my '1972 VW 1300 Beetle) to fix car problems (which, in essence, are real-life tangible problems).

LLMs can be used to create a search engine where the data used is actually a combination of your own data (like your own PDF repair manuals) and the information that Google has. With this combination, you should get more detailed information and repair instructions than Google provides for your car problems.

Typically, this combination is used to create knowledge bases, but in this case, we will check whether the results are actually usable.

English subtitles included!

Table of video contents
00:00	Intro
00:09	Can AI help with car repairs?
04:23	Tools used: Snowflake + Streamlit
07:16	Documents given to LLM+RAG
08:51	Use case 1: How does AI/LLM's solve this?
13:08	AI's answer to use case 1
19:19	Actual code used to the solve problem
22:15	Use case 2: How does AI/LLM's solve this?
23:24	AI's answer to use case 2
30:23	Summary