I will no longer engage in philosophical discussions about conscious AI/superintelligent machines, and here's why. (long\U0001f9f51/11)
— Giada Pistilli (@GiadaPistilli) May 27, 2022
This discussion between a Google engineer and their conversational AI model helped cause the engineer to believe the AI is becoming sentient, kick up an internal shitstorm and get suspended from his job. And it is absolutely insane. https://t.co/hGdwXMzQpX
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)
You May Also Like
To people who are under the impression that you can get rich quickly by working on an app, here are the stats for https://t.co/az8F12pf02
📈 ~12000 vistis
☑️ 109 transactions
💰 353€ profit (285 after tax)
I have spent 1.5 months on this app. You can make more $ in 2 days.
🤷♂️
I'm still happy that I launched a paid app bcs it involved extra work:
- backend for processing payments (+ permissions, webhooks, etc)
- integration with payment processor
- UI for license activation in Electron
- machine activation limit
- autoupdates
- mailgun emails
etc.
These things seemed super scary at first. I always thought it was way too much work and something would break. But I'm glad I persisted. So far the only problem I have is that mailgun is not delivering the license keys to certain domains like https://t.co/6Bqn0FUYXo etc. 👌
omg I just realized that me . com is an Apple domain, of course something wouldn't work with these dicks
📈 ~12000 vistis
☑️ 109 transactions
💰 353€ profit (285 after tax)
I have spent 1.5 months on this app. You can make more $ in 2 days.
🤷♂️
I'm still happy that I launched a paid app bcs it involved extra work:
- backend for processing payments (+ permissions, webhooks, etc)
- integration with payment processor
- UI for license activation in Electron
- machine activation limit
- autoupdates
- mailgun emails
etc.
These things seemed super scary at first. I always thought it was way too much work and something would break. But I'm glad I persisted. So far the only problem I have is that mailgun is not delivering the license keys to certain domains like https://t.co/6Bqn0FUYXo etc. 👌
omg I just realized that me . com is an Apple domain, of course something wouldn't work with these dicks