[kcdc 2025] AI Vulnerabilities in 2025 – Some of the darker sides of AI

Speaker: Andreas Erben

For more see the table of contents


From the news

What drives model behavior?

  • Initial training
  • Fine puning
  • Potential customer fine tuning
  • System prompt – how the model should behave.
  • Data/prompt
  • Filters on data, prompts, output

Other links

My take

After the first example, there was a bunch of content on how LLMs work. I tuned out a little during that section probably because familar. Then it got interesting again.

Leave a Reply

Your email address will not be published. Required fields are marked *