Discover more from unicorn chats
🦄 vol. 13
sash and claude chat about sustainable fashion and digital illusions
sustainable fashion ♻️
In 2019, I moved to Dublin and began working at my employer's EMEA HQ. Upon familiarising myself with the office, one of the first things I noticed was the "I work in tech" dress code: most people could be found walking around in jeans, t-shirts and sneakers. More specifically, I kept seeing a lot of Patagonia and a sneaker brand that I soon came to learn of, called VEJA. Anecdotally, I learnt that many of my colleagues wearing these brands were fans of the brands' stories, and the fact that they were (self-proclaimed) "sustainable".
What's sustainable fashion? 🧐
The fashion industry (as we know it) has become synonymous with the harming of the environment. Whether it's the manufacturing processes, the materials used, or the dumping of old / unsold products... The industry is no stranger to criticism (particularly fast fashion brands like Zara and H&M).
And while there’s no such thing as “eco-friendly clothing” (all pieces of clothing have at least some negative impact on the environment) there are a handful of brands focusing on combating various issues, for example:
Water wastage (using less water in the manufacturing process)
Hazardous chemicals (lessening the impact on community water sources)
Product life cycles (creating more durable clothing)
Waste (creating less trash and minimising dumping)
Agriculture (farming more organic materials)
The VEJA story 👟
The French sneaker brand was established in 2004, a little while after its co-founders (who were then working at an NPO) visited a handful of clothing factories in Asia, South America and Australia. The story goes that they were really shocked at what they saw.
"We realised that we did not know how products were made, and that if we knew, we would consume differently" said the co-founder.
Cue: launch a disruptive product in a highly saturated industry. 💥
Here are some of the ways that VEJA is pushing the boundaries:
They use (more expensive) wild rubber from from the Amazon and partner with local farmers
They use ecological cotton that's grown in a way that enriches the soil rather than damaging it with chemicals
They invented a recycled "bottle mesh" used to make the upper part of some of their sneaker models
1/3 of their shoes are 100% vegan
They voluntarily release an annual brand CO2 emissions report
They've begun offering a shoe repair service (in a bid to minimise product disposal)
Since launching, the brand has grown significantly and managed to achieve a reported $120 million in revenue in 2020, up from $78.5 million in 2019.
Zooming out 🤓
Of course, VEJA is not the only brand that is a known sustainability champion. Perhaps more widely known is Patagonia — a global brand best known for building strong customer loyalty. For almost 50 years, Patagonia has run its business with a strong stance on protecting and preserving the environment. Like VEJA, Patagonia is incredibly transparent about their products and processes.
Side note: Let’s keep an eye on the success of this IPO.
Bringing it closer to home 🌍
While there are a bunch of really exciting sustainably-focused brands on the continent, (as far as I know) there are none that have achieved the scale that the likes of VEJA, Patagonia and Allbirds have. Interestingly, these mentioned brands don't have a significant footprint on the continent either.
Personally, this raises a few questions:
Is sustainable fashion, and its less talked about virtue signalling, significantly important to consumers in our market?
What's stopping someone from creating and scaling the sustainable version of Bathu?
I'll let you decide for yourself.
digital illusions 🧐
Duck? Rabbit? It depends on how you are looking at it.
Chances are, you have come across this image at least once roaming the internet. This is the popular rabbit-duck illusion. It is a classic example of an ambiguous image, a visual form that has a fluctuating meaning to those perceiving it.
With the world looking more and more towards artificial intelligence for building intelligent computer vision systems, how does a machine fair when characterising this image?
The Google Cloud Vision API’s interpretation seems to depend on the rotation of the image. Looking at the various image orientations and the corresponding classifications, they kinda make sense. But can we make the model see something entirely different? 😈
Enter adversarial patches 🩹
Adversarial patches are inputs to deep learning models that are designed to make the model make a mistake. They do so by providing artificially constructed inputs that manipulate the output of deep learning-based computer vision systems.
There are different types of adversarial patches that are constructed slightly differently.
Imperceptible patches: Adding imperceptible noise to the input image (see Goodfellow et al). In this type of attack, you require the input image to create the adversarial patch.
Digital stickers: Adding a perceptible digital sticker to the input image (see Brown et al). This type of attack is particularly dangerous, because the patch can be crafted without knowledge of what the input image will be.
The implications? 😟
It is clear that the ability to manipulate the outcome of a computer vision system is dangerous. Let's look at some examples:
Self-driving cars - think of a malicious adversarial patch that makes the car "see" a green traffic light at a stop sign, or a stop sign in the middle of an intersection.
Facial recognition systems - these are becoming common-place in apartment blocks, airports. Adversarial patches could give someone unauthorised access to resources secured using facial recognition technology. It could also let people impersonate one another. This video for example shows a patch that makes the holder invisible to a classifier model.
The opportunity for exploitation is huge.
Can we fix this? 🔧
Defending against adversarial patches is an active field of research. A lot of the work on this has been focused on preprocessing the input images to deep learning models to remove the noise from the input images.
My stance is the same as when we talked about the implications of style transfer and deep fakes. The danger is real, but the dance between vulnerabilities and mitigations to them is also never-ending. As long as people are actively seeking out new ideas, it will continue.
here’s a cool article from claude, going over how adversarial patches can affect self-driving cars, and what’s being done to mitigate the risk
according to matt, all-in’s most recent episode gives a nice run down on crypto investing
karl’s back on his productivity tip with four truths, click on the link if you’re interested in optimising your time spent working