dshin

banned at 1 billion.
design @interlatent

[dshin @ddanielshin]

@bykahlil @Henryyinn @naturalpay @thinkwithmark @EasonTang23 so goated

[dshin @ddanielshin]

@camden_hulse @treenalabs @theresidency hell yea

[dshin @ddanielshin]

@sean_pixel our cto took a bite out of it

[dshin @ddanielshin]

@leylndd bro pick up your calls 😭

[dshin @ddanielshin]

@zjearbear @interlatent now here's a guy who knows ball

[dshin @ddanielshin]

@boulatbek @interlatent the future we’re building towards

[dshin @ddanielshin]

@leylndd can we gym together 👨‍🍼

[dshin @ddanielshin]

a simple guide for anyone interested in robotics. tune in for more

[Interlatent @interlatent]

Our mission is to make it easy for anyone to deploy a robot to help them in the real world

We wrote an intuitive guide to understanding modern robotics, catered toward an audience that understands technology but not AI robotics

We hope that this short blog post embeds in you the core principles that will bring further curiosity.

[dshin @ddanielshin]

@zjearbear @USC crossed this dude up in basketball

[dshin @ddanielshin]

@sypr3me work as in getting worked in 2k?

[seanpixel 🫧 @sean_pixel]

got MolmoAct2 working on @interlatent zero-shot on a random task no post-training

0% starvation with sub 50ms network latency and 130ms inference latency thanks to @allen_ai's custom cuda graphs (8B model btw)

we will provide GPU hours if you want to try

waitlist below

[dshin @ddanielshin]

@ashwinmukherjee this is sick

[dshin @ddanielshin]

@sean_pixel @interlatent can you bring me some toilet paper i’m stuck in the bathroom

[seanpixel 🫧 @sean_pixel]

we rented a 1000sqft studio in LA

to build out the entire physical AI deployment pipeline open-source @interlatent

We have the whole summer to research and implement the whole post-model stack:

inference / compute provision
automated evaluation
reward modeling
real-2-sim-2-real
RL in the real world
RL in a world model
observability

[seanpixel 🫧 @sean_pixel]

i own 0 A100s

but i can run cloud inference with sub 80ms network latency by renting one in Kansas (0% action queue starvation)

this is an early checkpoint of smolvla trained on 50 episodes for ~20k steps. now the goal is to improve on this with 0 expert demonstrations using RL

[dshin @ddanielshin]

@NorthstarRobots hell yea

[dshin @ddanielshin]

@WarnerTeddy lovely piece of writing, this is sick.

[dshin @ddanielshin]

@leylndd 😭😭

[dshin @ddanielshin]

@interlatent 🧘‍♂️

[Interlatent @interlatent]

Introducing our first technical blog post:
Automatic Dense Rewards for Autonomous Robot Learning

In this post, we utilize VLM chunking techniques to densely reward trajectories based on video and telemetry data without human supervision.