But frankly they don't need to because everything else you give them unthinkingly is way cheaper and way more powerful.
I'm back from a week at my mom's house and now I'm getting ads for her toothpaste brand, the brand I've been putting in my mouth for a week. We never talked about this brand or googled it or anything like that.
As a privacy tech worker, let me explain why this is happening. 🧵
But frankly they don't need to because everything else you give them unthinkingly is way cheaper and way more powerful.
Data aggregators pay to pull in data from EVERYWHERE. When I use my discount card at the grocery store? Every purchase? That's a dataset for sale.
If my phone is regularly in the same GPS location as another phone, they take note of that. They start reconstructing the web of people I'm in regular contact with.
Family. Friends. Coworkers.
To subliminally get me to start a conversation about, I don't know, fucking toothpaste.
It never needed to listen to me for this. It's just comparing aggregated metadata.
https://t.co/Hb7yiolQz3
Your data isn't just about you. It's about how it can be used against every person you know, and people you don't. To shape behavior unconsciously.
Block the fuck out of every app's ads. It's not just about you: your data reshapes the internet.
https://t.co/3ZCUELWqfA
At least make it hard for them. 🪡
Soon I\u2019m starting playtesting on my first adventure I intend to publish.
— Robert G. Reeve (@RobertGReeve) March 23, 2021
It\u2019s about a diamond cartel that\u2019s cornered the market on resurrection.
Here\u2019s the first piece of art I\u2019ve commissioned, by Andrew Phan. His portfolio is here: https://t.co/h3ScB5DbQg pic.twitter.com/0GkRhjuXXn
More from All
How can we use language supervision to learn better visual representations for robotics?
Introducing Voltron: Language-Driven Representation Learning for Robotics!
Paper: https://t.co/gIsRPtSjKz
Models: https://t.co/NOB3cpATYG
Evaluation: https://t.co/aOzQu95J8z
🧵👇(1 / 12)
Videos of humans performing everyday tasks (Something-Something-v2, Ego4D) offer a rich and diverse resource for learning representations for robotic manipulation.
Yet, an underused part of these datasets are the rich, natural language annotations accompanying each video. (2/12)
The Voltron framework offers a simple way to use language supervision to shape representation learning, building off of prior work in representations for robotics like MVP (https://t.co/Pb0mk9hb4i) and R3M (https://t.co/o2Fkc3fP0e).
The secret is *balance* (3/12)
Starting with a masked autoencoder over frames from these video clips, make a choice:
1) Condition on language and improve our ability to reconstruct the scene.
2) Generate language given the visual representation and improve our ability to describe what's happening. (4/12)
By trading off *conditioning* and *generation* we show that we can learn 1) better representations than prior methods, and 2) explicitly shape the balance of low and high-level features captured.
Why is the ability to shape this balance important? (5/12)
Introducing Voltron: Language-Driven Representation Learning for Robotics!
Paper: https://t.co/gIsRPtSjKz
Models: https://t.co/NOB3cpATYG
Evaluation: https://t.co/aOzQu95J8z
🧵👇(1 / 12)
Videos of humans performing everyday tasks (Something-Something-v2, Ego4D) offer a rich and diverse resource for learning representations for robotic manipulation.
Yet, an underused part of these datasets are the rich, natural language annotations accompanying each video. (2/12)
The Voltron framework offers a simple way to use language supervision to shape representation learning, building off of prior work in representations for robotics like MVP (https://t.co/Pb0mk9hb4i) and R3M (https://t.co/o2Fkc3fP0e).
The secret is *balance* (3/12)
Starting with a masked autoencoder over frames from these video clips, make a choice:
1) Condition on language and improve our ability to reconstruct the scene.
2) Generate language given the visual representation and improve our ability to describe what's happening. (4/12)
By trading off *conditioning* and *generation* we show that we can learn 1) better representations than prior methods, and 2) explicitly shape the balance of low and high-level features captured.
Why is the ability to shape this balance important? (5/12)