Gilberto Marcon

[Gilberto Marcon @gilbertoamarcon]

Really impressive, I'm exited to see this on a soccer competition.

[Skild AI @SkildAI]

We trained a robot to play football.

How? Self-play for 140 years in a virtual World Cup.

Meet the #Messinator:

[Gilberto Marcon @gilbertoamarcon]

Exciting to see this starting to take off. Tripling productivity overnight is proof this is bringing the AI revolution to the physical world.

[Skild AI @SkildAI]

We just hit 100M ARR within 10 months of starting deployments.

We are in factory lines. On construction sites. In kitchens. In data centers.

Cleaning. Welding. Building. Cooking.

Deploying.

[Gilberto Marcon @gilbertoamarcon]

This is the leap robotics needed. Instead of learning lots of tasks, this is learning the most fundamental physical interaction building blocks, learning how to interpret intent from a single demonstration, and learning how to assemble its building blocks to achieve that intent. That's a true generalist. Not learning tasks, but learning how to perform tasks from one demonstration.

[Skild AI @SkildAI]

Introducing S1, our new foundation model that learns from one example.

It can be taught 10-minute long tasks that it has never seen before, from one video prompt without any fine-tuning.

Watch S1 operate in real-time via in-context learning:

[Gilberto Marcon @gilbertoamarcon]

I think this is the breakthrough robotics has been missing. Instead of memorizing a large number of tasks, this is capturing the essential building blocks of physical interaction and the ability to parse demonstration intent. From one video demonstration of a new task, the model can distill the intent and use its physical interaction knowledge in novel ways to accomplish the intent.

[Gilberto Marcon @gilbertoamarcon]

@SkildAI I think this is finally cracking the code for robotics: Learning the generalizable building blocks of physical interaction and learning how to use these building blocks in novel ways to achieve the intent from a single demonstration.

[Skild AI @SkildAI]

Introducing S1, our new foundation model that learns from one example.

It can be taught 10-minute long tasks that it has never seen before, from one video prompt without any fine-tuning.

Watch S1 operate in real-time via in-context learning:

[Gilberto Marcon @gilbertoamarcon]

@kagan_tumer The Black Mirror episode "Hang the DJ", dramatizes evolutionary learning, although the simulation is unecessarily complex to be realistic.

[Kagan Tumer @kagan_tumer]

And while we're complaining about #AI depiction, let's acknowledge it when it is done right.

Please share any examples of accurate AI depiction in fiction that made you think: "hey, that's pretty close to how it might work."

[Gilberto Marcon @gilbertoamarcon]

I wouldn't say the AI depicted in #Westworld is unachievable, but I don't we getting it to be that general in the next few decades. https://t.co/ampm0cwYuW

[Kagan Tumer @kagan_tumer]

Many of us complain about how #AI is depicted in fiction/media. So, here's a challenge.

Please share any examples of poor AI depiction in fiction that made you think: "no way, that's not how AI works!"

[Gilberto Marcon @gilbertoamarcon]

@assfalk_ufg Professor, os requerimentos de desempenho serão passados durante a aula de terça?

[Gilberto Marcon @gilbertoamarcon]

@kleber_mc eu tava perdido aqui procurando o botão de like heheheheh

[Gilberto Marcon @gilbertoamarcon]

@assfalk_ufg professor, duas dúvidas sobre a Q1 desta prova: não estão invertidas as exp? A legenda está correta?