Posts
-
News with a snigger - courtesy of ChatGPT
I have taken inspiration from Joe Hovde and decided to restart this blog more like a journal than fully-rounded articles. So bear with me.
For a long time, I thought sarcasm was the ultimate test for an intelligent machine. You need to get some many things like context, references and timing to make a witty remark. We are now officially there. Chatgpt is pretty good at sarcasm. Some examples for you.
-
Can we make Test Cricket more fun?
Americans are amused at this to no end, but the original format of cricket, the Test Match, runs for a full 5 days. These long matches however, are seeing a slow decline over time, and the sport in recent times been dominated by shorter formats, but the 5-day Test still remains the highest echelon for greatness in cricket. I myself like Test Cricket over all other formats, but I have to admit that I simply don’t have the patience to watch a game slowly unfold over 5 days. Is there a way to make the viewing experience more fun and crisp? I give it a shot using a simple VGG Image Classifier here.
-
Choosing The Right Metrics (A Thought exercise)
Product success has several dimensions and it means different things to different people. It is important to choose the right set of proxy metrics to measure success and really think through how these numbers have the potential to mislead. Here I consider a hypothetical scenario where I am asked to measure the performance of a new recommender system on Twitter. This is how I would approach a set of metrics for this use case.
-
Building a reverse image search engine at EBTH
We built a reverse image search engine that looks up similar items sold by Everything But The House in the past, leveraging deep neural networks built on large image corpuses. Here’s how we went about finding the solution and implementing it.
-
Data Science and Overkill
Building machine learning models requires considerable amount of time, effort and skill. That much is well known. But what maybe not be obvious is that improving the performance of these models over and above a baseline is an exercise with greatly diminishing returns. Data science leaders would do well to understand the cost of iteratively fine-tuning these models and gauge it against business impact.
subscribe via RSS