Pitfall of (cheap) machine learning

Have you heard a story of how to kill trending video on youtube?

1. You subscribe thousands of blog accounts to channel you want to destroy.

2. After a video is published you unsubscribe all your bot's from the channel.

3. Youtube logic will assume it's because of the video and will remove a video from top lists.

4. Next day you slowly subscribe all your bot accounts to the same channel and repeat the process.


What about the story of Facebook helping spread fake information through a network?

1. Create a channel for small cats, funny jokes

2. Collect couple hundred thousands of followers

3. Change content from cats to fake news


Understand me correctly machine learning is awesome!

However cheap models could lead to problems.

To view or add a comment, sign in

More articles by Arturas Tutkus

  • Storage choose the right one

    In the book Fundamentals of Data Engineering, storage is one of the many covered topics as it plays a critical role in…

  • Do you know your data's ROI?

    In today's digital age, data is often seen as a valuable asset, and it's not uncommon for organizations to accumulate…

    1 Comment
  • "ETL is dead; long-live streams" is a false statement

    If someone in your organization is pushing for real-time processing for everything use this analogy: "We humans eat…

    18 Comments
  • Don't forget to migrate to GA4

    Some time ago Google has announced that Universal Analytics, also known as Google Analytics, will no longer process new…

  • Snowflake's UDF - array_like

    I'm working on a data model in which I have an array column. Long story short I need to select rows which would contain…

    7 Comments
  • MySQL 8 Window functions - best feature for me

    MySQL an open-source relational database management system has just released 8th version (Stable release: 8.0.

    1 Comment
  • Data engineer pipeline - from pixel to pixel

    Software engineers have term full stack. Maybe data engineers should have something similar? What about `pixel to…

Insights from the community

Others also viewed

Explore topics