Metal and glass recycling is great, aluminum recycling is about 95% less energy than mining new aluminum. Recycling glass is about 20% less energy than making new. These processes work and save energy and resources. Plastic does "pretend" recycling - Plastic recycling only actually exists to convince consumers that it's environmentally friendly, in practice the after-products of plastic recycling are significantly worse than the original, and only can be used in niche applications. The majority of plastic is now just forever garbage wherever it happens to end up. Pay the price of water and skies filled with plastic, or the cost of using less plastic.

So its just saying that Chrome tracks you 100% and now we are cutting off the ability of anyone ELSE to track you, so our tracking of you can be sold for a higher price. Any privacy discussion of this is just window dressing to sidestep the uneasiness you feel of Google knowing everything about you at all times.

The first one. They are tracking you through Chrome, so they will use their monopoly cut off any other ad tracking networks ability to track. This makes their tracking product relatively more unique in the market.

the Infrared Sensors on automatic toilets and sinks are calibrated for white skin. Facial recognition is the same. Poor performance on darker pigmentation is a result of poorly representative data sets and overlooking minority groups. When algorithms like these are used for real world decisions this has real-world negative consequences on minority groups. We can do better and we shouldn't accept the use of these algorithms when they perform poorly.

Deep Learning algorithms have gotten to the point where given enough data (billions of videos from hundreds of millions of people), understanding the factors that drive human attention is possible and moving from the obvious to the subtle.

They are looking into *your* brain even if you don't have TikTok because guess what, you are a human, and the drivers of your attention start from the same place as every other human. Sex, faces expressing emotion, places you've been, music you've heard, things that were popular in the years where you cared about pop culture, cute animals (which evolution already designed to hijack our love).

All these things are irresistible to humans in general, and when they are tailored to all the things your phone and advertisers know about you, result in you being unable to look away for an average of 5.2 videos. Pretty soon it will be 6.

Its not that Deep Learning algorithms are that smart, its that humans are trivial.

I have an uncle who was a Texas state representative. I don't think people realize that strict campaign finance laws are a federal thing, not a state thing. He was a rep for 6 years, and LITERALLY he had an ATT lobbyist who had a desk in his office, paid for all his meals, and even bought me and my cousin drinks when we were in Austin. When that lobbyist said "we've written this bill, sponsor it and submit it to the house" he did. Corporate capture of state governments and of the republican party at the state level is 100% complete.

I bet most people don't realize that when you are talking on the cell phone its not sending your voice. It compresses your audio based on a mathematical model of the human respiratory tract, sends over the few data points indicated by the model, and recreates the voice from this model on the other side.

It's close enough for phone calls, but doing with this with faces would save TONS of data, but I wonder how uncanny that valley really gets. Breaking down your face into component parts using a face model, just sending the model data, and then using deep learning to rebuild your face on the other side, you trade processing power for bandwidth.

At least when the technology is new you will probably look like a hellspawned demon, but eventually it will be standard and everyone will be used to it just like the slightly roboticized voice of conversing on a cell phone.

He walked us through his usual process for algorithm development. The cars have 7 cameras and the data science team at tesla is feature by feature adding things that can be seen using those cameras. It started with other cars and people, then lanes, then road vs. not road. Now they're working on streetlights and stop signs.

He is right Tesla uses real world data to improve the deep learning computer vision algorithms, but wrong basically about everything else.

"AI". Which in this case means machine learning algorithms, are not currently used for command and control of the car. That's done by traditional algorithms to regulate speed and position. The inputs to these algorithms however rely on deep learning algorithms to find the cars and lanes and keep that info updated in real-time using the cameras.

Sure, machine learning then.

Is that it will solve human behavior. It's not so difficult to model human behavior, and to affect it in meaningful ways. You don't need 100% accuracy, just affecting the behavior of 10 % of a population, userbase or customer group is enough to generate millions in profit, or votes, or views.

Another tool in the toolbox of the most powerful.

I interviewed a candidate who was an outgoing data scientist at Whole Foods, and she was there pre- and post- takeover by Amazon. She said that after Amazon came into the picture, Whole Foods got access, and took advantage of all of Amazon's data. If you swipe a credit card at Whole Foods that has been used on Amazon, the Whole Foods team gets your entire Amazon history, all your addresses and purchases for as long as you have had an account, including any location data they have collected on you from the Amazon app or any other apps they own. How likely do you think it is that Amazon also uses this information against its own employees? Almost certain that any applicable info from this data set is used to also track unionizing behavior or any sort of employee action.

Best summary ever. Thank you!

They have real challenges in making the jump from "Able to detect" to "Able to detect with high enough accuracy for self-driving". They are doing all the right things, and their ability to generate data from the fleet of cars (equipped with all the camera and sensor technology) is absolutely 100% unparalleled. However they are not there yet, and the way forward is still a research problem with no predictable end date.

Nope. HIPAA covers two groups: "Covered Entities" are healthcare providers (hospitals, clinics or doctors offices) and "Business Associates" are anyone the covered entities pays to deal with their data (medical billing, data storage, IT). Both are regulated under HIPAA and both can have access to fully identifiable patient information. If a covered entity wants to share information with a university researcher they would have to de-identify the data first, but they can share full data with anyone they have a BAA (Business Associate Agreement) with. HIPAA would still penalize Google or Microsoft if they leak any patient data, but other than that they can have access to every record the hospital system has. The caveat is they are only supposed to use the data for the business purpose laid out in the BAA, but there's not really any enforcement of that regulation, so it comes back to how much you trust those companies. HIPAA requires patient consent for sharing medical records with anyone but a business associate or other covered entity.

As a manager in tech right now I'm telling you Silicon Valley is not satire. All of the things from the show are basically day to day life at a startup. Tech startup life is an alternate universe.