Lincoln Berbert

Teaching robots to think like humans.

[Lincoln Berbert @Lincoln_Berbert]

@genex_games I should’ve read your op more carefully, you’re totally right. bruh now my optimization issues feel even more confusing lol

[Lincoln Berbert @Lincoln_Berbert]

@micro1_ai This was the same trajectory LLMs followed, robotics will be the same but with arguably higher demand.

[micro1 @micro1_ai]

As robotics foundation models improve, the highest-value training data will become increasingly specialized to specific tasks, environments, and edge cases.

Our latest paper breaks down how we’re building data pipelines based on this.

[Lincoln Berbert @Lincoln_Berbert]

@Rubzem What platform/harness are u messaging it through?

[Lincoln Berbert @Lincoln_Berbert]

@lightsilver323 The future of robotics is human:

https://t.co/ZHH301It0V

[Lincoln Berbert @Lincoln_Berbert]

@FrancisPSantora I set a 15-18 mo target with ability to cut costs and extend to 24 if needed

[Lincoln Berbert @Lincoln_Berbert]

@krgeorge Robots are part of the answer

[Lincoln Berbert @Lincoln_Berbert]

@sridharfyi We are raising!

[Lincoln Berbert @Lincoln_Berbert]

@TTrimoreau Yes

[Lincoln Berbert @Lincoln_Berbert]

@zdogmode I suppose everyone is wired differently but I don’t see it as days off. My business is “work” but it’s not “work.” I enjoy it.

That said sometimes I need to step back from the problem and gain perspective. A change of pace lets me digest information and attack from a new angle.

[Lincoln Berbert @Lincoln_Berbert]

@big_duca The way I’ve thought about it, is if you really care about the problem you’re solving, then you’ll be happy that it’s solved regardless of who solves it (because then you can move to the next problem)

If your only goal is to “get ahead,” then you’ll tear others down to do so.

[Lincoln Berbert @Lincoln_Berbert]

I’ve seen a lot of talk about how robots don’t have standardized evals, and while LLMs do have more robust benchmarks, LLM evals seem to have also scaled through custom rubrics & verifiers.

I believe the same approach will work in robotics as well.

[Lincoln Berbert @Lincoln_Berbert]

@starks_arq And yet people look at this and somehow go “this isn’t very good”

Like bro a literal rock just made this in under two hours. And that same rock will be smarter and better and more capable in less than a month.

Idk any human that can do that.. the fact that a rock can is insane.

[Lincoln Berbert @Lincoln_Berbert]

@thelcjones What about data?

not the typical “egocentric video” stuff but reward signals, rubrics, verifiers, human judgement and reasoning data that evaluates and determines quality?

[Lincoln Berbert @Lincoln_Berbert]

@AlexFinn I’m very bullish on voice as input.

[Lincoln Berbert @Lincoln_Berbert]

@anshuc Prompt?

[Lincoln Berbert @Lincoln_Berbert]

@brettcalhounn @redbudvc @Roots_Build You guys looking to invest in the next era of robotics data?

[Lincoln Berbert @Lincoln_Berbert]

@jackprice Yes, although it doesn’t feel like work.

[Adam Shuaib @adamshuaib]

Most extraordinary people I meet have an incredibly high tolerance for boring repetitive tasks. They are able to spend hours doing things normal people lose interest in within minutes.

We've seen many examples; a researcher who read the same paper 80 times trying to find a flaw, or a scientist who spent 17hrs repeating the same experiment hundreds of times before it worked, and a pianist who needed 20,000 playthroughs to perfect a certain Chopin piece.

We are falsely told by society to become well rounded and practise lots of different things, but great minds do the complete opposite; the 1st iteration is just an introduction, it's the 1000th iteration before you fully understand how something actually works, and virtually no one is willing to stay long enough to reach that point.