#MemoriesUpload - Twitter Hashtag

about 17 hours ago

@shawnshenjx #MemoriesUpload is huge internet scale first person video is exactly what robots need to learn how humans actually get things done!

0

35

Memories.ai

@memories_ai

about 22 hours ago

Robots learn by watching people work. #MemoriesUpload collects that video at scale: LUCI Pin to capture, our Large Visual Memory Model to index, and a pipeline built for 1M+ hours in 2 months. Frontier model labs are already training models with the data. Read more on the blog 👇

1

4

0

2

219

AGTP

@AGTPinsights

2 days ago

https://t.co/mEbpRHiGeJ just launched Memories Upload, a network that pays people worldwide to film everyday work for robot training. Here's what you need to know. CEO Shawn Shen announced it today with #MemoriesUpload. His point: robots need internet-scale first-person video of how people get things done. Collecting that at high quality takes hardware, model and infra working together, and the company says it spent over a year building all three. People record their daily tasks with wearable cameras. Mounts include headband, cap, chest and wrist. The gear is under 45 grams and has a camera and a light. Footage uploads on its own, and every accepted hour earns pay, with higher rates for the hands-on tasks robotics labs need most. The material does not give exact pay rates. Then the footage gets indexed. https://t.co/mEbpRHiGeJ says it is filling a video data lake across every task, environment and hardware setup. Its Large Visual Memory Model indexes each clip into hand pose, head pose, object mesh, action segments and dense captions. Each hand is tracked and labeled separately, frame by frame. Labs can search the data by what each hand is doing. Shen calls it not a labeling tool, but a world indexer. The demand is already there. https://t.co/mEbpRHiGeJ says 3 of the top 5 frontier robotics labs are training models with the data. Commentators say the real bottleneck for robotics is clean, diverse video of humans doing real work, not models. Others say scaled, automated first-person data pipelines are replacing manual teleoperation datasets. Key numbers: - 3 of the top 5 frontier robotics labs already training on the data - Over 1 million hours of first-person video recorded so far - Goal of more than 10 million hours - 1K+ tasks across 20+ industries - Wearable camera under 45 grams - Shen's launch post passed 103K views within hours EMG force data is listed as coming soon, so the dataset is set to grow beyond video.

AGTPinsights's tweet photo. https://t.co/mEbpRHiGeJ just launched Memories Upload, a network that pays people worldwide to film everyday work for robot training. Here's what you need to know.

CEO Shawn Shen announced it today with #MemoriesUpload. His point: robots need internet-scale first-person video of how people get things done. Collecting that at high quality takes hardware, model and infra working together, and the company says it spent over a year building all three.

People record their daily tasks with wearable cameras. Mounts include headband, cap, chest and wrist. The gear is under 45 grams and has a camera and a light. Footage uploads on its own, and every accepted hour earns pay, with higher rates for the hands-on tasks robotics labs need most. The material does not give exact pay rates.

Then the footage gets indexed. https://t.co/mEbpRHiGeJ says it is filling a video data lake across every task, environment and hardware setup. Its Large Visual Memory Model indexes each clip into hand pose, head pose, object mesh, action segments and dense captions. Each hand is tracked and labeled separately, frame by frame. Labs can search the data by what each hand is doing. Shen calls it not a labeling tool, but a world indexer.

The demand is already there. https://t.co/mEbpRHiGeJ says 3 of the top 5 frontier robotics labs are training models with the data. Commentators say the real bottleneck for robotics is clean, diverse video of humans doing real work, not models. Others say scaled, automated first-person data pipelines are replacing manual teleoperation datasets.

Key numbers:
- 3 of the top 5 frontier robotics labs already training on the data
- Over 1 million hours of first-person video recorded so far
- Goal of more than 10 million hours
- 1K+ tasks across 20+ industries
- Wearable camera under 45 grams
- Shen's launch post passed 103K views within hours

EMG force data is listed as coming soon, so the dataset is set to grow beyond video.

Shawn Shen

@shawnshenjx

2 days ago

Today we launch #MemoriesUpload. Robots need internet-scale first-person video of how people get things done. Collecting this at high quality takes hardware, model and infra working together. That's what we've spent over a year building. And already, 3 of the top 5 frontier robotics labs are training models with the data.

76

1K

148

393

1M

2

1

0

1

346

Praveen Kumar Verma

@Alacritic_Super

2 days ago

Robots need real-world, first-person video to actually learn how people use their hands instead of just lab simulations. Awesome to see @shawnshenjx and the team launch #MemoriesUpload with their lightweight wearable camera and hand-tracking data, already used by 3 of the top 5 robotics labs.

0

4

1

359

Memories.ai

@memories_ai

2 days ago

Physical intelligence needs to learn from the real world. That's what we're building with #MemoriesUpload. We turn first-person footage of everyday human work into robotics training data at scale. Already, 3 of the 5 frontier robotics labs are training models with our data. A big thank you to Wen Shao at @FortuneMagazine for covering the launch.

memories_ai's tweet photo. Physical intelligence needs to learn from the real world. That's what we're building with #MemoriesUpload.

We turn first-person footage of everyday human work into robotics training data at scale.

Already, 3 of the 5 frontier robotics labs are training models with our data.

A big thank you to Wen Shao at @FortuneMagazine for covering the launch.

1

5

0

2

606

Shawn Shen

@shawnshenjx

2 days ago

Today, we announced #MemoriesUpload, collecting and indexing internet-scale egocentric video to accelerate physical intelligence. 3 of the top 5 frontier robotics labs are already training models with the data. Thanks to Wen Shao at @FortuneMagazine for covering the launch. It’s a big moment for us and for the future of physical AI.

shawnshenjx's tweet photo. Today, we announced #MemoriesUpload, collecting and indexing internet-scale egocentric video to accelerate physical intelligence.

3 of the top 5 frontier robotics labs are already training models with the data.

Thanks to Wen Shao at @FortuneMagazine for covering the launch. It’s a big moment for us and for the future of physical AI.

4

58

15

9

3K

Memories.ai

@memories_ai

2 days ago

The next leap in robotics will depend on learning from the real world. We've spent over a year building the video models and infrastructure to make this possible. #MemoriesUpload

Shawn Shen

@shawnshenjx

2 days ago

Today we launch #MemoriesUpload. Robots need internet-scale first-person video of how people get things done. Collecting this at high quality takes hardware, model and infra working together. That's what we've spent over a year building. And already, 3 of the top 5 frontier robotics labs are training models with the data.

76

1K

148

393

1M

7

13

1

4K

Shawn Shen

@shawnshenjx

2 days ago

Today we launch #MemoriesUpload. Robots need internet-scale first-person video of how people get things done. Collecting this at high quality takes hardware, model and infra working together. That's what we've spent over a year building. And already, 3 of the top 5 frontier robotics labs are training models with the data.

76

1K

148

393

1M

Top Tweets for #MemoriesUpload

Last Seen Hashtags on Sotwe

Trends for you

Most Popular Users