Published:
Previously, we explored how artificial intelligence is currently in a transition from static algorithms to continuous Perception-Action Loops. We observed this first in the physical space with robotics and then in digital execution environments with multi-agent systems.
Published:
In my previous blogs on the “The Rise of Physical AI” and “Agentic AI”, we explored how artificial intelligence is currently in a transition from static algorithms to continuous Perception-Action Loops. We observed this first in the physical space with robotics and then in digital execution environments with multi-agent systems.
Published:
In my previous post, on “The Rise of Physical AI”, we explored how the robotics industry is currently in a transition from static algorithms to continuous Perception-Action Loops, allowing autonomous systems to perceive, reason and act in real-world environments.
Published:
After spending time working with multispectral computer vision payloads and embedded systems, it became apparent how different it is to process data on a screen versus the physical, real-world interaction. For the past decade, Artificial Intelligence has been strictly “disemmbodied”, where it lives inside servers in Data Center, processing text and generating pixels on a screen. However, the next frontier of machine learning isn’t just about thinking, it’s about acting especially in the real, physical world.
Published:
After taking a deep dive into the computational infrastucture demands of modern Artificial Intelligence, I realised that while our models are scaling exponentially, the physical infrastructure required to run them is hitting a massive bottleneck. We tend to trust and assume that these so called “Hyperscale” data centers to process everything from every day user LLM query prompts to Autonomous Vehicle Networks assuming that our current power grids infrastructure can just endlessly accomodates them out of the box.
Published:
After taking an Advanced Artifical Intelligence course, I realised that while modern AI models achieve very highly in terms of benchmark performance, they have a massive, invisible blind spot. We tend to trust these so called “Black Box” systems with increasingly critical taks from Self-Driving Vehicles to Autonomous Robotics and Medical Diagnosis - assuming that if they have a high accuracy, they are also “intelligent”.
Published:
In 2017, Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uzkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin published the Research Paper “Attention Is All You Need”, which introduced the Transformer model architecture, but is Attention really all you need? In this post, I’ll summarise the key ideas in this Research Paper and share my takeaways of whether or not “Is Attention Really All You Need”?
Published:
In this post, I built a Simple Neural Network for XOR made in Python from Scratch without any external Machine Learning Libraries/Frameworks such as PyTorch or Tensorflow or JAX.