📈 Do it quick, then do it right.
Good morning everyone, well at least the European readers among you! 🇪🇺
I am writing this just after finishing my hockey match and it seems I have come down with a cold. Playing and umpiring in the rain certainly didn’t help, I can tell you that much. 🏑 🌧️
Anyway, enough of my complaining, let’s get into this week’s edition!
Thanks for reading Dishing The Data! Subscribe for free to receive new posts and support my work.
In 2020, I undertook a research year as part of my master’s at the National Physical Laboratory in London. Besides the pandemic, it was quite an experience and ultimately culminated in my decision not to pursue physics research, but that is by the by.
My first supervisor at the time was (is) an acclaimed academic in the field of metrology, Michael De Podesta. Literally on my first day, Michael gave me some of the best advice that I still use now:
“Do it quick. Then, do it right.”
It may seem simple, but it’s a really powerful message.
We should just figure out how to do something, and then worry about how we can make it better. In other words, be action-oriented and start before you are ready.
I apply this thought process when implementing anything new in my data science projects. I get the thing working, then start making it ‘nice and tidy.’ You learn more by doing the work, instead of falling into the planning fallacy.
This is the same approach I have taken with my YouTube channel. At the moment, my goal is to make 100 videos and get better each time. I don’t spend hours trying to get every video “right,” but focus on getting them done, then slowly iterate.
Of course, this rule doesn’t apply to every field. A heart surgeon can’t just “do it quick.” However, if your application has no catastrophic downside, this framework will greatly help you, particularly on problems you may not be familiar with.
What’s Been Cooking 🥘
Some tasty stories this week:
AI Reads Herculaneum Scroll - A 21-year-old computer science student created a machine learning algorithm that has managed to read a burnt Herculaneum scroll from the 79AD Mount Vesuvius eruption.
New IBM AI Chip - IBM’s new NorthPole processor is specifically designed to run neural networks which are the backbone algorithm of things such as ChatGPT and DALL-E.
Baidu Releases Ernie 4.0 - Chinese tech company Baidu releases their new AI model Ernie 4.0, which they claim can rival OpenAI’s GPT4. This is another installment in the US-China technology battle.
Ukraine’s AI “Killer Drones” - Ukraine is using drones that utilise AI to find and “neutralise” targets without any human intervention. We literally have “killer robots.”
Weekly Favourites ❤️
📚 Book - The Snakehead by Patrick Radden Keefe. Earlier this year I read Keefe’s recent book about the Sackler family and Oxycontin. This one is just as thrilling and is about the gangs, smuggling and illegal activities inside New York’s Chinatown.
📖 Course - Video Editing with Final Cut Pro X. It’s probably no secret that my video editing style is inspired by Ali Abdaal. I have been taking his amazing Final Cut Pro course on Skillshare and have received quite good feedback regarding my own videos!
📝 Blog - Activation Functions & Non-Linearity. This is my second post on my journey of neural networks. This one delves into the reason we need our networks to be non-linear and how activation functions achieve this.
(PS: Some links are affiliate links that I get a kickback from with no extra cost to you 😎)