Level up your understanding of Gen AI this week in just 66 minutes
or less if you like to listen on double speed
I’ve been snacking on the new Generative AI for Everyone course from Andrew Ng this week. I’m a long term Andrew fan because of his ability to make very technical subjects accessible to a wide audience. For well over a decade, he has been a reliable resource both for those who want to learn the basics themselves and for those who need great reference material to recommend to those around them keen to upskill. If you’re already fairly comfortable with the foundations of Generative AI and find what follows too basic for you, then don’t overlook this second point! I lean on Andrew all the time to bootstrap the knowledge of teams and stakeholders. Borrow with pride!
In week one of the course, you (or your loved ones / stakeholders) get an overview of how LLMs work and an accessible explanation of why many people are calling AI a ‘general purpose technology’ akin to the internet or even electricity (as Andrew has been saying for a while now but frankly seems a bit of a stretch to me at the moment).
Andrew suggests using Gen AI for
writing tasks e.g. as a brainstorming tool or for draft copy
reading tasks e.g. for summarisation of long texts or classification of emails from customers
chatting tasks (e.g. company specific chat interfaces for services, hopefully ones that are actually useful this time around)
He also has some helpful advice for how not to build a chatbot that does terrible things accidentally (which has been a bit of an unfortunate theme recently).
And true to form, Andrew has a new, easy to remember rule of thumb / mental model for figuring out at high level, whether the task you are considering is one that a standard ‘plain vanilla’ LLM is likely to help with today without additional augmentation.
The above are just three takeaways that caught my ear, as I listened at 1.5 times speed and made good use of pause and rewind. Take a look/listen and distribute wisely to ease your upskilling burden.
In other news that caught my eye this week
Apparently Midjourney thinks men are mostly over 60, while women are rarely over 40. Which seems both odd and biologically unlikely.
And in an evocative illustration of how curly the problem of removing bias can be
OpenAI found that when it filtered training data for its DALL-E 2 image (to weed out pornographic or violent images generator, it exacerbated gender bias. In a blog post, the company explained that more images of women than men had been filtered out of its training data, likely because more images of women were found to be sexualized. As a result, the data set ended up including more men, leading to more men appearing in results.
How AI reduces the world to stereotypes, restofworld.org
Great food for thought and absolutely worth a browse. One of the cases in which pictures convey far more than words ever could.
And finally a cautionary tale about the unintended consequences of a rush to automate away all forms of human interaction (apologies, paywall but the below will give you the gist).
We have two big supermarket chains in Australia. Both have understandably been quick to roll out automated checkouts … and like many other grocery retailers, have struggled with ‘shrinkage’ as some folk play fast and loose with the task of swiping their own groceries through in a ‘reliable and high fidelity' manner.
However it turns out that combating the dishonesty of a few with tactics that are humiliating to the many - most recently video cameras providing live feeds of you as you swipe your groceries - is something to deploy at your own peril , particularly when you have just removed a customer contact point where employees might have the chance to understand and assuage customer dissatisfaction.
Be thoughtful and curious before you disintermediate your own human employees too far from your customers with cost saving automation. When AI changes a task in a workflow, odd things have a habit of happening to the overall workflow itself.
While you may not categorise self scan checkout technology as AI, the trust challenges experienced by the big brand supermarkets relying on it highlights the need to take a holistic approach to systems change when you are thinking about deploying AI automation at any point in a customer journey.
Wishing you sunshine for your weekend, or snow if that’s your hemisphere. I have an 18th birthday party to prepare for (no, not mine) so wish me good weather.