How LLMs can make stuff up and what to do about it — Main Objectives Working with large language models is not without risks including responses based on what’s called a LLM “hallucination.” Hallucinations can be a serious problem for LLMs because they can lead to the spread of misinformation, expose confidential information, and create unrealistic expectations about what LLMs can do. Understanding hallucinations and…

We’re really just dating, but it’s going pretty well — What’s Google Bard? I’m going to let Bard tell you… Google Bard is a large language model, also known as a conversational AI or chatbot, created by Google AI. It is trained on a massive amount of text data, and is able to communicate and generate human-like text in response to a wide…

Garbage in, garbage out has never been more true. — Rise of the Machines Large language models (LLMs) are able to conversationally provide an unprecedented level of ML-generated information when asked the right question. For example, LLMs can generate articles (no, an LLM didn’t generate this text), poetry, song lyrics, have a conversation with someone about a topic, interpret some types of documents /…

SQL + ML = magic What’s BQML? One of the less intuitive parts of building machine learning models is sorting and selecting data (e.g., using Pandas). Wouldn’t it be nice if you could just use SQL? Yes, it is nice that with Google Cloud Platform (GCP) BigQuery ML (BQML), you can combine…

What’s the Point? — One of the most fundamental concepts in data science involves the idea of a “prediction” (the word should probably be “inference”). But how does that happen? What’s going on under the covers? What’s THE fundamental concept? To be thorough, the answer is, “it depends,” but to avoid an overly cerebral…

Use y as the starting point for X — good for machine learning — What’s the Problem? When dealing with a dataset with many features, it can be difficult to separately define the features (X) and the target(s) (y). Defining the target(s) is usually straightforward because there is usually one variable in the target (or there are very few). …

It’s a shapeshifter I say! — Basic Confusion (Matrix) When predicting a categorical variable target, an important model metric tool is the confusion matrix, and that tool is aptly named because the confusion matrix can be very confusing (that’s not even why they call it a confusion matrix). However, with a simple tip, you can avoid confusion and let…

A Surprisingly Tricky Bit for Beginners — One of my professional lives is as an adjunct professor teaching undergrad data science. In that capacity, in predictive analytics (which is a basic overview of data science for all intents and purposes), I’ve found a consistent and surprising pattern with students: They struggle to understand the differences between regression…

Even before you get to the math! — Data science is hard, but not because any single concept is especially hard (noting that some are). The hard part is keeping the basics straight because: Synonym soup Obvious subtlety No clear answers A million situations The basics aren’t basic Synonym Soup It’s hard enough to keep the relatively vast number of…