Outlier Labs

A builder salon for top technical talent exploring non-obvious research directions.

[nat @natal1e_c]

using two real SO-101 pick-and-lift demonstrations as a starting point, we trained a residual policy in MuJoCo with EGGROLL. 19 GPU min, 30,720 rollouts, 0 safety rejections. next: real-world rollouts to close the sim-to-real gap. #robotics @ask_joon @bidiptas13 @outlierlabshq

[Outlier Labs @outlierlabshq]

@natal1e_c @ask_joon @bidiptas13 Wow this is super cool @bidiptas13 ! Thanks for coming to our event!

[Outlier Labs @outlierlabshq]

@HedeyaY @bidiptas13 Let’s goo

[Omar Hedeya @HedeyaY]

🚀 Two days left for the EGGROLL challenge!

​​EGGROLL tutorial: https://t.co/7uLnbuF85K

To submit: post what you did tagging @outlierlabshq and @bidiptas13

​​The winner will be featured by Bidipta & go for dinner with the Eggroll team in September.

[Omar Hedeya @HedeyaY]

@outlierlabshq @bidiptas13 You can find tutorial here: https://t.co/7uLnbuF85K

[Omar Hedeya @HedeyaY]

So here is our challenge for you!

Build something creative with EGGROLL, tag @outlierlabshq and @bidiptas13 before 26th of August.

Get featured by Bidipta and win a dinner with the Eggroll team in London! 3/n

[Omar Hedeya @HedeyaY]

For the first time you can actually train networks with non-differentiable activation functions, you can train on full harness, you might even be able to unlock higher reasoning capabilities unreachable by RL 2/n

[Omar Hedeya @HedeyaY]

What if we could train LLM at the speed of inference with less compute?

We had a pleasure of hosting @bidiptas13, co-inventor of EGGROLL algorithm, to talk about the potential of evolutions strategies. 1/n

[Outlier Labs @outlierlabshq]

@HedeyaY @bidiptas13 Excited for this!!

[Bidipta Sarkar @bidiptas13]

Build and share a project using EGGROLL this month!

[Outlier Labs @outlierlabshq]

@ask_joon @bidiptas13 Thank you for joining our first edition! Can’t wait for what you build with Eggroll 🥚

[Outlier Labs @outlierlabshq]

@DGneusheva @bidiptas13 @HedeyaY Can’t wait for what everyone builds with EGGROLL 🥚

[Daria Gneusheva @DGneusheva]

What if we could train LLM at the speed of inference with less compute?

We had a pleasure of hosting @bidiptas13, co-inventor of EGGROLL algorithm, to talk about the potential of evolutions strategies.

Build something creative with EGGROLL and share on X, tagging @outlierlabshq and @bidiptas13 before 26th of August.

Get featured by Bidipta and win a dinner with the Eggroll team!

Link to EGGROLL tutorial in the comments.

[Daria Gneusheva @DGneusheva]

What if backpropagation isn't the only way to train AI? Could we train models at the speed of inference?

Join us on the 30th of July in London for the first edition of Outlier Labs with @bidiptas13, PhD at Oxford and co-inventor of Eggroll algorithm.

We'll work through toy examples together, then spend the rest of the afternoon hacking and experimenting on where Eggroll could lead us.

If that sounds like your kind of afternoon, we'd love to have you join us.

Link to join in comments.

@HedeyaY