Shapath Das

Predicting next actions | ceo @unsupervizedai
Ex-Resident @lossfunk ex-EIR @AntlerIndia

alumni @TelAvivUni
ex @WorldCinemaFund

[Shapath Das @shapathdas]

Introducing PRM (Physical Reasoning Model): a foundation model that infers and outputs an uncertainty-calibrated physical state from robot experience, so downstream policies can train on physics-complete signals — not just pixels and actions.

@unsupervizedai @Ranita_Jana

[Shapath Das @shapathdas]

@babugi28 Congratulations!

[Shapath Das @shapathdas]

@babugi28 Looking forward!

[Shapath Das @shapathdas]

@narsagna congratulations!

[Shapath Das @shapathdas]

@xeepak @ChatGPTapp Good suggestions Deepak, also these days the answers are taking longer inference time compared to @claudeai

[Shapath Das @shapathdas]

@after_ephemera @partabot_ Yes, indeed, they delivered the bot a week late.

[Shapath Das @shapathdas]

@partabot_ Hey what is the best way to get in touch with you?

[Raj Patel @babugi28]

who wants bounding boxes and segmentations? we’re launching next week, and researchers in academia are eligible to receive our datasets for free. drop a comment and i’ll dm you

[Shapath Das @shapathdas]

This concept bot has so may issues! Why it even has two legs why not wheel based!

[Ronald van Loon @Ronald_vanLoon]

Dual-Arm Bipedal #Robot Designed for High-Payload Logistics Operations
by @tweetciiiim

#EmergingTech #Technology #Innovation #Tech

[Shapath Das @shapathdas]

@ShubhamKukretii @Visa Really cool! So so happy for you both!

[Shapath Das @shapathdas]

@AnselDias @norma_core_dev Super cool!

[Shapath Das @shapathdas]

Hi @danfei_xu just finished reading your Article titled "To summon a Sensorimotor Ghost", really insightful, thought I should invite you to read my take on the data problem.

[Shapath Das @shapathdas]

https://t.co/qCzifUKJd4

[Shapath Das @shapathdas]

Danfei and the EgoMoma team needs a clear shoutout!

[Danfei Xu @danfei_xu]

https://t.co/Y1jBFTAJSf

[Shapath Das @shapathdas]

Humans have evolved an intuitive understanding of physical laws—not by learning from scientific books, but through rich, multisensory experience of the world.

Now we want machines to acquire a similar understanding purely from video data. Is it working?

https://t.co/uOykaLGrTg

[Shapath Das @shapathdas]

https://t.co/qCzifUKJd4

[Shapath Das @shapathdas]

@mattfreed Thank you for sharing @mattfreed

[Shapath Das @shapathdas]

Hey folks,

Is anyone traveling to Bangalore from San Francisco this week?

[Shapath Das @shapathdas]

“Strategy is product, and product is strategy”

— Jan Simon Veicht

[Shapath Das @shapathdas]

“What once felt magical now felt inefficient. What was left were nostalgic memories of the good old days, which in hindsight weren’t that good. That’s how paradigm shifts work.”

A few best lines from the @Alfred_Lin essay!

[Alfred Lin @Alfred_Lin]

https://t.co/d31hFe7w3O

[Shapath Das @shapathdas]

Robotics isn’t compute-constrained.
It’s data-constrained.
Specifically: #teleoperation_data.

[Shapath Das @shapathdas]

Teleoperation is to robotics what labeled data was to deep learning -- scarce, expensive, and ultimately the defining bottleneck.

[Shapath Das @shapathdas]

Compare that to AI

LLMs → trillions of tokens (free-ish)

Vision → billions of images (internet-scale)

Robotics → ~10⁵–10⁶ trajectories (painfully collected)

We’re 6–9 orders of magnitude behind.