Blog

Latest Post

Feb 11, 2024

Analysing LLM Mistakes

One of my main frustrations with LLMs is the way they make mistakes. To me, the seemingly random insertion of incorrect or hallucinated code fundamentally contradict how I usally read and think about code. For example, reacently I tried to find all the red sections in an image with some Python code, and I had a rough idea of how it should work. An image is essentially and array of pixels and each pixel is a set of numbers representing it’s color. A format called RGB is commonly used for this, where each pixel consist of three number in the range of 0-255, representing how much red, green and blue light make up the pixel. You loop over the array and check whether the color values of a pixel fall with a range that would be red, meaning it should have high values for red, and lower values for green and blue.

Read more

All Posts

2024

  • Feb 11, 2024 - Analysing LLM Mistakes