I'm Adam Bronte, a long time software engineer based out of San Diego, CA. Currently I'm building an AI agent for HR and people teams at Paradigm. Recen...
LLM benchmarks are sort of an odd thing. Every time a new model is released these fancy graphs saying they are X% better in Y category and compare themselves against each other. Some times they line up to real world usage, but most of the time you should probably just ignore them. (Including this one! probably) Introducing Wordle Bench . A simple benchmark on how well a model can solve wordle.…
I like to read and I read a good amount. When I'm reading a series and I'm waiting for the next book to come out, I often forget what happened in the previous book. This seems like a great use case to build a rag! What is needed? A book in epub An epub parser Chunker Vector store LLM to query A book in epub There are tons of public domain ebooks you can get started with at Project Gutenberg .For…
You should use the product you are building, yes? It seems obvious but there is a bit of nuance to this process. While building Corgea , we try to use our own product as much as possible. We use its core functionalities and its integrated into a normal flow. The other day I was using Corgea to accomplish a very specific task. There was a different intent around my usage and brought a new…
Whenever I'm trying out a new LLM, my go to prompt is always Write a hello world program in python. It's really simple and you can get a sense for how the LLM responds to such a simple request. How many different ways are there really to write hello world? Take GPT-4o for example: Certainly! Here's a simple "Hello, World!" program in Python: print ( "Hello, World!" ) To run this program, you can…
When building Fresh Tracks I needed to put together a dataset of a bunch of ski resorts and some information around them. In the past, this would of been a tedious task either trying to scrape the data together or maybe you could find something already existing. With todays LLMs, this becomes really easy to do. It feels like such an obvious, yet non-obvious use case LLMs can solve. The only thing…
Hallucination in the LLM world simply means the model made a bad prediction. The model generates a token, or series of tokens that doesn't have any real world viability or doesn't relate to your original input. I'd also consider any result that explicitly against your instructions a form a hallucination as well. One example I came across recently was when I was asking GPT-4o to give me some…
With Kamal 2.0 the ability to deploy non-ruby and non-rails apps was added. I've been using Kamal to deploy a couple of Rails apps and its been great to work with. I wanted to try this out for some other non-rails projects and thought I'd share what I found since this isn't a well documented use case. Lets say I have a python background job that prints out the current time every second. import…
A while back some colleagues and I were talking about what if you could flood the internet with specific content, could you influence today's popular LLMs? Pretty much all LLM's today are trained on publically accessible data on the internet. One of the primary sources of that data comes from Common Crawl. This is a free and open-source dataset of a large portion of the internet, spanning more…