The one about UX, user cohorts and GPT 3
Welcome to another edition of Product Principles, the newsletter where I bring the best links related to tech and product while sharing some of my recent learnings. I've recently started posting product inspiration and tips on Instagram and Twitter . I made a super handy guide for interviewing users as well. You can get it for 50% off with promocode ‘newsletter’
I usually joke that Product Management is all about managing opposites.
Look at the tree, but also look at the forest…
Plan a strategy, but also execute…
… understand the logical side of business and technology, but also understand UX+UI
So naturally I was thrilled when Ethan, one of our subscribers, asked me a great question a couple of weeks back:
“How do you ensure UX Design is incorporated into product teams and as a discipline more broadly at a company?”
I thought this was a solid question and one that definitely warrants more scrutiny now more than ever as design, engineering and product teams are geographically apart. What used to happen next to a whiteboard now happens over Zoom.
Engineering and design culture are a direct reflection of company culture. When both planning and execution are centralized it's common to see words like “monolith architecture” and “design teams” pop up.
Centralized planning, decentralized execution. If the company and product values are well aligned between stake holders we usually end up discussing “how” more than “what”. You're more likely to end up discussing “the best way of conveying value to the user” rather than “what should we do?”
If the company is product driven, favours ux excellence and the PM has done a good job in understanding the problem: it's now an execution + iteration issue. This is ideal. 🚀
UX emphasis is prioritized by the founding team's culture. As the company matures it stops becomes less “vision” and more “objective+metric supported”. If your company is going through growing pains: don't forget to scale design.
In the end of the day Design+UX is part of the Product team. They should be aligned and prioritize accordingly. Culture and metrics should cross pollinate over time. If the Design and Product teams are not in lockstep: trouble incoming.
Best resources for PMs to keep up with UX best practices:
Don't Make Me Think
Design of Everyday Things
amazon affiliate links present
📘 Product stuff
Adjacent User theory (30 mins)
This is a long article and arguably my favorite one of the month. Bangaly Kaba (ex Instagram, Facebook and Instacart) talks about his model for cohorting and pushing users to cross the engagement gap. Bangaly's Adjacent User Theory is a handy model to help you understand your user distribution and help you push them towards increased engagement.
I also really liked his advice about building personas based on behaviour instead of demographics. Give it a read, it's a really good one.
Why OKRS might not work for your company ( 7 mins)
For every “use this framework” article there is a mirror article somewhere on the internet explaining why it might not be such a good idea. Marty Cagan of Inspired breaks down important aspects of company culture that must also be present for OKRs to work. Spoiler: Feature teams vs Product teams are completely different creatures.
Those successful companies aren’t successful because they use OKR’s. They use OKR’s because it is designed to leverage the empowered product team model. And as I have tried to make clear with years of articles and talks, the empowered product team model is a fundamentally different approach to building and running a tech-product organization.
The company isn't successful because it uses OKRs, but instead it's successful AND uses OKRs. Think of OKRs a multiplier for good company product culture.
📰 Interesting news and articles:
GPT 3: Why it's more exciting than Bitcoin (and scary as hell)
A few years back Elon Musk got together with his friends and started OpenAI an artificial intelligence lab. The third instalment, GPT 3 came out earlier this month and… it's very hard not to get hyped up about it. 😊 Here's why:
Machine learning algorithms are usually split into two types:
Supervised: “Hey computer, here are 1 million photos of chairs. Hopefully by the end you'll understand what a chair is and isn't.”
Unsupervised: “Hey computer, here's a load of data, see if you can group them into recognisable clusters. Try and find me some similarities.”
GPT 3 is an unsupervised algorithm with ~175 billion parameters that can perform specific tasks straight out of the box. As a user you can type in your request and it will attempt to understand and act on it.
Here's some really interesting things that it was spotted doing out in the wild:
Understood natural text (plain english) and turned it into a SQL query
A Figma plugin where the person just described (again, in plain english) what they wanted the app to look like.. and it drew it perfectly!
Wrote poetry pretending to be Richard Feynman copying the style of Robert Frost
Was asked what makes bread fluffy. It then understood on it's own what fluffy meant and found a paragraph on a Wikipedia article it felt would answer it accordingly. Fluffy was nowhere present on the text.
Where GPT3 is exciting:
Besides being exciting new tech it's essentially a great way of building a first version anything. It's all the fun + creative parts of coding without the grind. Its capacity of understanding language is phenomenal and breaks down previously required knowledge barriers. Maybe at some point we'll be building our MVPs in plain english.
Where I think it stumbles:
When testing the viability and strength of AI/Machine learning models we usually end up hearing terms like “Turing Test” being thrown around. A Turing test is essentially a fancy way of saying “Could this algorithm fool me into thinking I was interacting with a human?”. One person with beta access to GPT 3 decided to put it to the test.
In summary: It works really well in the beginning when asked questions that Google could answer. It understood very well what is a more likely human answer (“I walked my dog” vs “I walked my banana”) but failed to identify what was absurd. Ask it silly or nonsensical questions and it will look like a computer trying to act cool.
Where it get's scary:
It's an issue of manipulation and weaponisation of tech.
In 1969 with humanity's first moon landing a backup speech was prepared for President Nixon in case the mission failed. The text is publicly available in the internet but fortunately it was never delivered. Some very bright individuals at MIT built a perfect Deepfake model of Nixon delivering the speech. It's a hauntingly realistic video of an alternate universe. What happens when Deepfake and GPT 3 meet? Can't we just submit any kind of content and have a deepfake model act accordingly? How do we know what we're seeing is real?
…or better yet: Imagine you're a college student with a 2,000 word paper due in 2 hours. What's to stop you from just feeding GPT 3 the essay prompt and hitting enter?
The Slack Social Network 20 mins:
This is a great article that tackles the power of default and bundling. Microsoft Teams is a Slack competitor on the surface but in reality it's a free alternative designed to keep you locked in the Office 365 ecosystem. Over 36 million users later and it's clear it's meant to retain not acquire new clients. My favorite quote:
“This is what Slack — and Silicon Valley, generally — failed to understand about Microsoft’s competitive advantage: the company doesn’t win just because it bundles, or because it has a superior ground game. By virtue of doing everything, even if mediocrely, the company is providing a whole that is greater than the sum of its parts, particularly for the non-tech workers that are in fact most of the market. Slack may have infused its chat client with love, but chatting is a means to an end, and Microsoft often seems like the only enterprise company that understands that.”
If I were Product at Microsoft I would be doubling down on building a Microsoft Teams developer ecosystem.
Benedict Evans’ presentation about changes in behaviour caused by Covid.
It gets extra interesting after ~slide 26 where you start to see crazy movements in consumer behaviour. Everything from unemployment rate by industry to VR demand.
The Twitter hack
Earlier this month several prominent Twitter accounts were taken over with the same message:
Three very strange things about this hack:
- Evidence showed willing and active participation of Twitter employees. This is not the first time that a Twitter employee has been caught acting outside of the company's best interest. Late last year we had a Twitter employee charged with spying for Saudi Arabia
- After hijacking such prominent twitter accounts… why go for such a crude tweet? Asking people to transfer their bitcoins over looks lazy. If I were a conspiracy advocate I'd argue that it was all a demo or a diversion.
- All tweets pointed to the same Bitcoin wallet. You can actually monitor all the money flowing in and out of it over here. Looks like a pretty big paper trail
😎 Cool things from the web
Lego and Nintendo teamed up and built the ultimate nostalgia kit.
Costs almost as much an actual console but damn is it impressive.
Google's UK search trends
`running` searches increasing followed by `knee injury` 😛
This newsletter is a labour of love. I cherish user feedback and would love to hear what you liked/would like to see improved on the next edition.
If this newsletter was helpful or entertaining in any way, don't forget to hit the share button below :)
Have a great Friday!