roboticsPundit's profile picture. Robotics Pundit. 
Should be trusted as much as a regular pundit.
Interested in the intersection of Robotics/Product/Research

t.will

@roboticsPundit

Robotics Pundit. Should be trusted as much as a regular pundit. Interested in the intersection of Robotics/Product/Research

You might like
t.will reposted

For my product friends, if you want to work at a fast-growing company, double-digit ARR, and a category leader, one of my portfolio companies is hiring. DM and I'll share more.

oanaolt's tweet image. For my product friends, if you want to work at a fast-growing company, double-digit ARR, and a category leader, one of my portfolio companies is hiring. DM and I'll share more.

t.will reposted

In the last couple of days I discussed multiple times about how I am using AI Agents to develop some pet projects faster. Would you be interested to watch a video series where I share my experience with practical use cases in robotics? I will do it if I get 100+ likes 😉


Worth a look

Most VLAs have no memory. MemER is a scalable method for giving VLAs long-term visual memory. Key ideas: - relevant info is in a small # of frames: store just a few key frames - manage & use memory at high level of abstraction, with a HL policy Paper: jen-pan.github.io/memer/



Progress towards … what? Do these people know what a product is?

New: Across Silicon Valley, AI researchers are putting in 80-100 hour workweeks to keep pace with technological progress. Several compared it to war, and one executive even joked that the new work schedule is "0-0-2," midnight to midnight with two hours off.

MeghanBobrowsky's tweet image. New: Across Silicon Valley, AI researchers are putting in 80-100 hour workweeks to keep pace with technological progress.

Several compared it to war, and one executive even joked that the new work schedule is "0-0-2," midnight to midnight with two hours off.


t.will reposted

"the human brain doesn't need tons of training data to do stuff"

francoisfleuret's tweet image. "the human brain doesn't need tons of training data to do stuff"

t.will reposted

#PlotJuggler 3.12 is out, with the first significant change in the UI in a long time. github.com/facontidavide/…


This is true, and also a great encapsulation of what drives modern robotics development. [everything but product + talking to customers] 😅😬

Robotics is everything you could wish for working in engineering. Mechanical, electronics, machine learning, maths, physics, creativity, art, all coming together to form a system you can hold in your hands.



t.will reposted

There is no such thing as "trans ideology." The most prominent transgender person in the country is a lifelong Republican, Trump donor and former Olympian legend. The difference between left and right on this is not indoctrination, it's acceptance.


Starbucks lying again!

roboticsPundit's tweet image. Starbucks lying again!

This structure of tweet is always confusing. Defending the guy in the CCCP shirt with technical war over land minutia means you lost the argument already. Why? It’s attention. Both messages have the primary theme [Russia attacking EU]

The US could not hold Iraq (a desert) with ~500,000 soldiers in-country at peak. Assertions that Russia (a country w/a demographic problem already) are going to overrun Europe (an area 23x larger than Iraq, with far more challenging terrain) simply do not pass the smell test.



🤔 🤷🏼‍♂️ Pickpocketing is driven by tourism because thieves do not have to steal from their neighbors.

Pickpocketing requires delicacy and skill honed over years of practice Our petty thieves are lazy, they don’t have to work for it. Jumping turnstiles and running out of stores with armloads of household goods in broad daylight. No finesse Good times create weak thieves



An anonymized retweet! 🎉 Time to retire! @grok - do you think @roboticsPundit is a bot?

Pickpocketing requires delicacy and skill honed over years of practice Our petty thieves are lazy, they don’t have to work for it. Jumping turnstiles and running out of stores with armloads of household goods in broad daylight. No finesse Good times create weak thieves



A platonic Frank Luntz tweet. Clever as long as, you ignore the context and don’t think for more than 2 seconds

Biggest news story of this weekend.



30 minutes on Twitter goes a long way towards demonstrating “tech’s” lack of critical thinking 😓

💀💀💀💀💀 it’s happening already (sent to me by a friend)

JayElHarris's tweet image. 💀💀💀💀💀 it’s happening already (sent to me by a friend)


Walk at human speeds. 🤷🏼‍♂️

The sim2real RL pipeline we’ve built scales to many different challenging motions. This tech will soon be extended to useful manipulation tasks.



🤔 This is a really great point about the parallels of AI and fiction in that soon no one will believe this is AI because it’s too crazy to be. AI will not be able to generate moments that are earnestly unbelievable.

This is just never not going to look AI generated



Just a great piece of modern art. 🙏

It's day 12 of my TAPBOT side quest where we're teaching impressionable AI models to doomscroll. I've got a pretty solid swipe going now. Next I'm going to use @moondreamai to tap likes on my Following tab or scroll if there aren't any unliked posts.



Everyone’s opinion doesn’t matter. The positive signal is that they are talking about it. Is it worth doing? Does it deliver value? Don’t do hard stuff because it’s hard. Do meaningful stuff because it’s meaningful.

everyone saying it can't be done is the positive signal means the challenge is grand, the reward is massive, and actually… it's the only thing worth doing



🚨 🚨 🚨 Achieved: tweet with 50 likes Initiate transition to full time farmer

Defining characteristic of staff engs is clarity and understanding how the sausage gets made. Discussing that with junior engineers is rewarding, but with leadership and product and security, it gets exhausting. Better to find a hard place where their failures are their own.



t.will reposted

This robot data alone is not enough! A couple key improvements First: the model training recipe Compared to the π-0 recipe (flow matching): We find far better performance when pre-training with next-token prediction (i.e. π-0 fast recipe) and fine-tuning with flow-matching.

chelseabfinn's tweet image. This robot data alone is not enough! A couple key improvements

First: the model training recipe

Compared to the π-0 recipe (flow matching):
We find far better performance when pre-training with next-token prediction (i.e. π-0 fast recipe) and fine-tuning with flow-matching.

United States Trends

Loading...

Something went wrong.


Something went wrong.