LLM interview Question — Mention Key failure points in a Retrieval Augmented Generation (RAG)

Amber Ivanna Trujillo
3 min readApr 27, 2024

Alright, let’s dive deep into the murky waters of Retrieval Augmented Generation (RAG) systems and explore the treacherous territory of failure points. Brace yourselves for a rollercoaster ride through the highs and lows of AI mishaps, illustrated with a dash of humor and a sprinkle of technical wizardry. Feast your eyes on this masterpiece of an illustration, adapted from a renowned research paper that breaks it all down in a visual feast!

  1. Missing Content: Picture this: the retrieval strategy pulls a disappearing act, leaving the LLM scratching its digital head and replying with a classic “I don’t know” or worse, conjuring up some wild hallucinations. Open-source example? Think of instances where your Google search turns up nada, zilch, zero useful results.
  2. Missed Top Ranked: Ever had that sinking feeling when the answer you seek is buried deep within the database, but your trusty RAG system fails to unearth it? It’s like searching for treasure and finding a rusty old nail instead of a shiny gold coin. Check out blogs on improving search algorithms for some insights.
  3. Not in Context: You’ve got the documents in your grasp, but somehow they slip through the cracks and fail to make it into the LLM’s cozy context bubble. It’s like having…

--

--

Amber Ivanna Trujillo
Amber Ivanna Trujillo

Written by Amber Ivanna Trujillo

I am Executive Data Science Manager. Interested in Deep Learning, LLM, Startup, AI-Influencer, Technical stuff, Interviews and much more!!!