Blog posts

Decoding the Long Context Craze: Is It Worth The Hype?

TL;DR: In the wild world of LLM providers, there’s this race going on about having the largest context possible. The big question: Is it a game-changer or just a money-making gimmick? Here’s my two cents – it’s pitched as a top-notch feature, but from where I stand, the reality might not match the hype. Back to Basics What is “context”? The context is the information that is given to the model for it to generate text. Read more →

December 4, 2023

Navigating Networking as an Introvert: 10 Effective Tips

You are an introvert? You feel awkward among people you don't know or crowds? I have tested a strategy for you! Read more →

November 14, 2023

LLMs: Determinism & Randomness

While working with GPT-3.5 Turbo and GPT-4 models, I expected that the temperature parameter would allow the models to have a deterministic behavior. But very consistent is actually a better characterization. I ran some experiments to see how temperature and top_p parameters impact the models' output — in practice. My conclusion on this: when building LLM-powered apps, embrace the randomness! Read more →

July 28, 2023

Using ChatGPT to build a Gitlab Merge Request reviewer with GPT

I tried a merge request reviewer using OpenAI model with a simple prompt. The final code is available here: https://github.com/malywut/gpt_mr_reviewer Read more →

January 21, 2023

Hand drawn diagrams in PowerPoint

I discovered the “draw” menu in PowerPoint, as it magically appeared when I plugged in my pen tablet. By combining it with the “sketch” shape outline, I made nice diagrams with a hand drawn feel. Read more →

December 12, 2022