CLOUDY podcast | #34 From Language Models to Humanoid Robots

What is the fundamental difference between a classic industrial robot and a humanoid robot?

The classic industrial robot has been with us for nearly 50 years. It is essentially a robotic arm in a factory that picks up one component and places it onto another. Over the years, its precision has increased to tenths of a millimeter, but it still lacks legs—it remains stationary. A humanoid robot has legs and can move, and to achieve this, it had to be trained specifically using AI. It is effectively a "semi-human" that brings artificial intelligence into physical hardware.

What was the main impulse for acquiring a humanoid robot and starting its training?

I have been working in the field of AI for a long time, and I see a growing trend where it is great to actually "touch" the AI. ChatGPT or Copilot live in your computer or phone, but a robot is physical hardware into which you can transfer that knowledge. The first impulse was a trip to China, where we saw AI deployed in everyday life. The second was a specific project for an automated store, SOR 24/7. Imagine a grocery container where a robot, upon request, walks to a shelf, picks up a soda or a beer, and brings it to the customer. That is a real-world use case we are currently working on.

What is the primary purpose of the Senti robot, and where can we meet it today?

The primary purpose is retail work, where it operates in a safe, glass-enclosed environment. Additionally, we use it in education. Senti has visited schools where children—from first graders to high school seniors—watched how robotics works. The robot also serves as a massive marketing magnet. At conferences like AI Day in Bratislava, people naturally gravitate toward it, take photos, and show interest in the technology. At that specific conference, Senti even delivered its own keynote speech, speaking and moving for 13 minutes, which was incredibly engaging for the audience.

How does the training and programming of such a humanoid robot work?

When you buy this hardware, it’s like a blank laptop. It only knows the basics—how to walk and wave. We teach it to communicate by connecting it to cloud services and Large Language Models (LLMs). Setting up the basic connection between the microphone, speaker, and AI takes a few hours or days. Teaching physical skills is more demanding. For that, we use virtual simulators (digital twins). If we want the robot to pick up a glass, it practices the movement a million times overnight in the simulation. Once it learns it in the virtual environment, we upload the resulting model into its "brain," and it then repeats the action in the real world without error.

Can a robot be a solution for loneliness among seniors or a helper in healthcare?

Definitely. Today we can talk to a phone, but having a physical partner next to you is a completely different experience. For seniors or bedridden individuals, a robot can be a companion that not only communicates but also monitors. It can remind them to take medication, measure blood pressure via connected devices, or call a doctor or family if the person feels unwell. While it can’t put socks on a senior just yet (which is technically very difficult), in the near future, it will be a helper that brings a glass of water from the kitchen or fetches a phone.

When will humanoid robots become a common part of our daily lives?

We are currently in a phase I would compare to ChatGPT 2.0. The hardware is already at a high level; now we are waiting for the software leaps. I anticipate that within a horizon of two to five years, we will encounter them much more frequently. They will first establish themselves in industry and retail—places with labor shortages, such as rural shops or for routine factory tasks like unloading boxes onto a conveyor belt. Subsequently, they will start appearing more in offices at receptions, and eventually in our homes.

You can listen to the full podcast on 👉Spotify, 👉Apple Podcasts, or watch it on

👉YouTube.