Avi Bhargava

Something always going on

[Dhawal Jain @thatssodhawal]

people who rely on coffee to push through their day:

[Dhawal Jain @thatssodhawal]

JUST IN: User Interview #4

@VeronicaBuron, founder of Fluffle.

Like any founder, she's constantly pulled in many directions at work, which often left her feeling scattered and overwhelmed.

She turned to Mave because she wanted her focus back, without the crash that comes with coffee.

Now, 78% of her days start with Mave.

https://t.co/F1fP3tc7dB

[Dhawal Jain @thatssodhawal]

Maverick interview #2

Don't let biology hold you back

Order today at https://t.co/B6LhpMn058

[amv @aryanmadhaverma]

my learnings from reading the paper if anyone is interested
https://t.co/OYf8GgdNYZ

[amv @aryanmadhaverma]

molmoact2 has really strong zero shot spatial understanding but i found it to be lacking in terms of last mile grasp for objects which might not be in its training distribution.

tested out the policy with random objects in my room. an inhaler i bought from thailand would definitely be OOD for the model, but given a prompt "pick up the object and place it on the coaster", it was able to identify the vague object and how to reach it.

it failed to grasp it every time. (it was able to grasp a redbull can though)

i finetuned the action head and froze the backbone though it is recommended to let the gradients pass through the VLM.

the arm started picking the object up. the stutter you see in the video is because I ended up using a synchronous loop which waits for the entire chunk before executing. synchronous because i was painfully bottlenecked by latency. i thought a rented L4 gpu in europe would allow me to temporally ensemble subsequent action chunks but the queue drained before the next chunk arrived. will try this with a local gpu soon. faced some compatibility issues with lerobot as well so ended up creating my own inference API.

also kinda reinforces that we're still early for cheap and reliable deployments and a lot of exciting problems are yet to be solved :)

[Aman Kumar @amankrsingh03]

Nothing on our Mave Headset is accidental.

Every curve, every edge, every material choice was taken after countless discussions, early user feedback, and way too many simulations. tbh, I learned more about Ansys simulations for this than I did in 4 years of college.
For example: our electrodes. Everyone said go rectangular or circular, it's the standard, easier to manufacture.
But, feedback and simulations told us otherwise.

This is what building hardware that actually works looks like, why hardware is actually hard.

[Avi Bhargava @curious_fork11]

@thatssodhawal Can’t wait !!!

[Dhawal Jain @thatssodhawal]

Many people asked why the headset looks the way it does, dropping a video on Wednesday to tell you everything :)

[Avi Bhargava @curious_fork11]

@regalstreak @dakshpixelup Bwahaha

[Avi Bhargava @curious_fork11]

@jaisharmaxyz @Forbes @mavehealth Let’s fucking goooo boss!!

[Jai Sharma @jaisharmaxyz]

Made it to the @Forbes 30U30 Asia list

Coincidentally, today also marks 3 years since I left Newton School to build Mave.

@mavehealth has been one of the most exciting and defining decisions of my life so far.

It’s surreal to think that we have built a device that helps 100s of people across the globe manage stress and improve their productivity.

Forbes has always been a bucket list item and couldn’t have done it without @thatssodhawal , @amankrsingh03 and the amazing team at Mave.

[Dhawal Jain @thatssodhawal]

Made the @forbes 30u30.

Over the last few months, we scaled @mavehealth from a prototype to an actual headset that hundreds of people use to regulate stress and focus better.

Watching it become part of their daily routine, something they actually look forward to, has been the most surreal payoff.

Grateful to be building this with the best team. Day 1 energy, always.

@forbesunder30

[Avi Bhargava @curious_fork11]

@venturetwins @ponnappa Gemini models think wayyy too much. A simple decision even causes anxiety in the models. Overthinking in humans leads to the same outcomes that this town-simulation yielded

[Shimorekato @iam_shimorekato]

@GiorgiaMeloni Melody company founder after watching this

[Avi Bhargava @curious_fork11]

@vatsal_sanghvi Bhai, itne jargon dosti ko describe karne mein thodi use karte hai! Friendship just happens.

You wanna hangout at your favourite dosa joint?

[Avi Bhargava @curious_fork11]

@thatssodhawal The age of polymaths and savants is here! The world will be shaped by people who believe specialization is for insects, not humans!

Hope the same reckoning hits the people running orgs as well!

[Avi Bhargava @curious_fork11]

@dbzkunalss @regalstreak bring your a game! What a chance to grab!

[Dhawal Jain @thatssodhawal]

Get a chance to train with UFC fighter Max Griffin.

- 2 hour MMA training (in San Francisco)
- No prior experience expected, fun vibes only
- It’ll be a memory you cherish for life

To participate, repost the post below and tag @MaxPainGriffin in the comments.

Max will randomly pick 20 people tonight.

You’ll hear back from us if he picks you.

[Avi Bhargava @curious_fork11]

@monali_dambre Models are just the brain. Would love to know if you have made breakthroughs on harness and specifically on tool use costs via OSS models!

[Avi Bhargava @curious_fork11]

@thatssodhawal What an anthem! Saving money to get my own Mave real soon!

[Avi Bhargava @curious_fork11]

Woahhhhh! @MaxPainGriffin is here!

[Dhawal Jain @thatssodhawal]

Introducing Maverick #1

Max "Pain" Griffin, UFC fighter

- started using Mave headset 100 days ago
- for stress regulation and performance

Brain score: 71 (day 0) -> 83.4 (day 30)

Grab your headset: https://t.co/B6LhpMmsfA