🍫 MSA: MARS Surveillance Agency

Vending machines are spying on you, NVIDIA to the Moon, and why package tracking texts are sketchy

Author

Brady
February 26th, 2024

In today’s issue:

  • An error spurred a student to discover vending machines on campus are using facial recognition

  • A look into NVIDIA’s earnings and the future of AI Computing

  • Why those package tracking texts are easy phishing targets

  • A tool with the the only AI product in its market

First time reading? Subscribe here!

'Facial recognition' error message on vending machine sparks concern at Canadian University

A student at the University of Waterloo in Canada noticed an odd error on a vending machine referencing facial recognition. Looking into it more, they realized vending machines owned by MARS Candy and manufactured by Invenda were collecting data on students using the vending machine.

Photo of the error that led to the discovery of the facial recognition software

Invenda said it collects and processes demographic data about individuals using the machine, like gender, age, etc. They claim to do this locally, and the data is used in targeted advertising, including with advertising "partners." In a marketing brochure for the vending machine, you can see the data they collect and how you can connect it to their "Internet of Retail" platform.

They want you to know it is "based on IoT, AI, and cloud technology," as mentioned in the brochure many times. This vending machine is somehow GDPR complaint, but the student who discovered this problem questions whether its legal Canada. It gives big "let's put radio on the internet" vibes.

Imagine all this tech and data collection to sell you a pack of gum or a candy bar for $2. How can they squeeze this further in the future?

If customers who buy these machines and place them in locations want more money, advertising isn't a bad extra source of income. If you look at these machines as real estate, they can be billboards in relatively targeted locations without the need to collect personal information.

Redbox did something similar, with some sites having a screen showing ads. Other vending machine manufacturers have built ad systems, and most of them collect data, but is that necessary? Maybe we can put little servers in vending machines and sell decentralized computing power instead. Imagine a future where every vending machine has a computer mining Bitcoin.

NVIDIA Announced 4th Quarter Earnings, More Than Tripling Sales

Demand is so high, and NVIDIA's H100 chips are so hard to get; companies are rationing usage of GPU computing power like its war. Google and Microsoft have "rationing" committees that help distribute computing resources to internal departments and external customers via their respective cloud platforms. Most big tech companies have various projects and products that need these resources.

NVIDIA can't meet demand primarily because they contract out the manufacturing to Taiwan Semiconductor Manufacturing Company (TSMC), which has limited manufacturing resources and has other customers like Apple, AMD, and Qualcomm, all of whom also benefit from the AI boom. TSMC has several ongoing construction projects for building new fabrication plants worldwide, but most are years out. Competitors like AMD are far from being on the levels of NVIDIA's GPUs, and even if they could catch up in compute power, NVIDIA has almost vendor lock-in with their software suite, CUDA.

One place where we can see a shift, though, is how a more settled AI market can shift how computing power is utilized in data centers, where most of the GPUs these companies own are located. When creating an AI model, training is the most compute-intensive part, where NVIDIA's H100s mostly come into play. During NVIDIA's earnings call, CFO Colette Kress said that more than 40% of the their data center business in the past year was for deploying AI models, better known as "inference."

Inference deals with tasks like generating responses from the model, which requires significantly less computing power and does not need to be done on more powerful H100 chips. This trend is where other chip companies like Intel can come into play, where deploying the model for inference makes more economic sense on a more power-efficient and less powerful computing platform. That was a hit at Intel, but it's also true: most competitors are far behind NVIDIA and their 80% market share for AI chips. Some startups are popping up to develop chips, but they will likely get to the scale of the big players later.

NVIDIA will continue to dominate and likely also benefit from the next big technology boom. Gaming, Cloud Computing, Cryptocurrency, AI, what's next?

Interesting Reads

Cool Finds

  • DiskClick

    • Do you ever want to hear old hard disk sounds as a sort of weird, nostalgic throwback? Well, this program lets you do just that, based on actual disk usage on your C drive.

  • once.tools

    • Subscription fatigue is real. Once aggregates different software that you only need to pay for once, no subscription is needed. It has a small library, but it was just launched Sunday the 25th.

  • Collov AI

    • AI has made its way into every industry, even interior design. Collov AI is a platform that turns your interior pictures, into a style of your choosing. I would specifically like to shout out their cabinet design AI where Collov boasts how they have the “only platform for AI-driven, vendor-sourced cabinet rendering”. Cant imagine too many other companies competing with them right now.

Thinking about daily driving Linux again. What do you think?

Have a good week!