Reducing Hallucinations by 60% Without Changing the Model
When you build an AI product on top of a third-party LLM, you don’t control the model weights. You control everything around it. That constraint turns out to be more of an advantage than a limitation.
Measuring AI Impact When You Can't A/B Test
If you ship fast enough, you’ll eventually outrun your ability to measure. That’s exactly what happened with Slack AI.
Some Toy Algorithms - Sentiment Classification
This is an attempt to do a sentiment classification for the movies. We will build a binary linear classifier that reads movie reviews, as you would see from the sites like Rotten Tomatoes, and predicts whether the result is positive or negative