Salman

Training robotics foundation models & accelerating scientific discovery; prev cofounder @omegalabsai and AI @WOMBO 🇨🇦/🇵🇰

[Salman @_sshahid_]

Our work on the @StanfordSVL BEHAVIOR Challenge revealed something unexpected: less reactive control modes performed significantly better.

Committing to 50-step plans (vs replanning every timestep) boosted success from 30% → 100% on subtasks like navigating & grasping.

[paul pavel @polpvl]

just learned that YC released a last-minute AI application process that tells you if you get an interview in 10s

https://t.co/cXttgDEahK

[Salman @_sshahid_]

New solution to the sim2real gap: upload your self into sim thereby eliminating the gap entirely.

[Salman @_sshahid_]

Taste is the last frontier

[Datapoint AI @datapointai]

1/ Introducing Datapoint AI

Intelligence is being commoditized. Good judgement is still rare.

We collect human preference data that trains frontier models, by replacing mobile ads with preference tasks.

[jai @jai_mansukhanii]

We just raised $7.2M USD to make insurance feel magical.

Nobody has ever said "my insurer was so easy to deal with."

We're changing that.

@generalmagic_ai builds AI agents that text your customers and complete insurance workflows end to end.

30 minute quotes now take 3.

[Salman @_sshahid_]

@EricHasegawa1 Excellent space!

[Salman @_sshahid_]

@allenzren That makes a lot of sense, thank you. Maybe we generate new embodiments in sim and unlock a lot of cross-embodiment data that way. Saw some interesting papers at neurips on automated design of new embodiments

[Salman @_sshahid_]

@LihanZha Awesome, thanks, would be cool indeed

[Salman @_sshahid_]

This is incredible work: a recipe for training VLAs that can generalize zero-shot to unseen embodiments. Open code and weights too

[Lihan Zha @LihanZha]

Today's state-of-the-art VLAs struggle to generalize zero-shot to new robot embodiments, despite training on extensive multi-embodiment data.

We introduce Language-Action Pre-training (LAP) and LAP-3B — the first VLA to achieve substantial zero-shot transfer to unseen real-world robot embodiments, through simply aligning action representation with language.

Everything is open-sourced! Try it out on your own robot:
🌐 https://t.co/dWuDtbV0Tn

[Salman @_sshahid_]

@LihanZha Of course, and thank you for the excellent work! I was curious: do you have an intuition on how to scale LAP to other form factors like humanoids, quadrupeds etc? Higher dimensional EE pose vectors maybe?

[Salman @_sshahid_]

@deedydas 1X

[jai @jai_mansukhanii]

In 1990, a team left Apple with ideas the world was not ready for.

They did not win commercially, but their thesis held true.

We’re doing the same for insurance, using language to cut the friction between intent and outcome.

Today, OpenSesame is becoming General Magic

@generalmagic_ai

[Salman @_sshahid_]

@silennai Great points on when to use Cursor vs when Claude Code makes more sense

[Silen Naihin @silennai]

After 5 years of coding with AI, I've written a comprehensive guide.

The only limit should be your own ideas: https://t.co/n6ZyRbOSPD

[Salman @_sshahid_]

@jacobrintamaki Would love to be added!

[Salman @_sshahid_]

@k7agar Yep 😭

[Salman @_sshahid_]

We live in incredible times 🫶

[Eddy Xu @eddybuild]

today, build ai is open sourcing the largest egocentric dataset in the world

- 100,000 hours
- 14,228 workers
- 10,800,000,000 frames

each pixel is a unique video (try below)

[Salman @_sshahid_]

Incredible work by Ilia and team!

[Ilia @IliaLarchenko]

A couple of days ago, I presented our 1st place solution for the 2025 BEHAVIOR Challenge at @NeurIPSConf . Now, we've open-sourced our solution: code, model weights, and a detailed tech report.
Let me unpack what we did 👇

[Salman @_sshahid_]

@ziqiao_ma Would love to meet up in the Bay! Are you in town for the Humanoids Summit?

(Your DMs aren’t open btw)

[Salman @_sshahid_]

Someone needs to make a credit card that earns extra rewards on $$$ spent on GPU credits