rkoush

Do AI models cheat?

Sun, 17 May 2026 00:00:00 +0000

Reward Hacking

Do AI systems cheat or always perform what the intended task is? In this post we explore an idea where AI can exhibit sycophantic behavior or user pleasing behavior.

GPU architecture

Fri, 01 May 2026 00:00:00 +0000

GPU

GPU or graphics processing unit, are type of processors that are used to perform large floating point instructions. It gained intial popularity for it’s use in video processing or gaming and more recently for training large language models.

But why are GPUs so fast and are they always faster than CPUs? The answer lies in the architecture. A GPU is designed to perform single instruction over multiple data acronymed SIMD. A CPU usually contains Control Unit, Arithmatic Logic Unit, registers etc. More recently CPU have multiple cores, each cpu works like a mini cpu with it’s own CU and ALU. If a CPU has 4 cores, it can operate 4 instruction truly in parallel. GPU are different, they can 1000s of threads in parallel.

Some quotes

Fri, 01 May 2026 00:00:00 +0000

My agenda

Give more than you take in work and in relationships
Work for what you want and enjoy the process
Have morals and respect for yourself and for others
Do great things, do fun things, do learn
Relax and listen to music
Be kind to others but first to yourself

TPU? Google's answer to Nvidia

Fri, 01 May 2026 00:00:00 +0000

Tensor processing Units (TPU)

Why is AI Safety Challenging?

Fri, 01 May 2026 00:00:00 +0000

Safety for weak capability?

We are postulating that in near future AI capabilities will grow so much, that it will be superior to human intelligence. In such a reality, enforcing some mechanism that allows AI to still work for the progress of humanity will be necessary. I have been reading the book “Superintelligence” and an opinion that exists among researchers and thinkers, is that we have once chance to build that level of AI that surpasses human intelligence and when we get there, if we don’t do it in a safe way, that could possibly mean in worst case end of humanity.

About

Mon, 01 Jan 0001 00:00:00 +0000

The Journey: From Scaling Systems to Safeguarding Intelligence

I am Rohit Kaushik, a Software Engineer exploring the intersection of distributed systems, AI alignment and interpretability.

Professional Odyssey

My career has been a pursuit of understanding complex systems, both digital and metaphysical.

The Zurich Era: I completed my Master’s in Zurich, where I honed my foundational understanding of rigorous engineering and research amidst the precision of Swiss innovation.
The AWS Years: Post graduating from my Masters, since the last 4 years, I have been serving as an SDE at Amazon Web Services (AWS). Working at the “Cloud Scale” taught me how to build for millions of concurrent users, managing the entropy of distributed systems where a single millisecond of latency or a minor edge case matters.
The Pivot to AI Safety: I was recently accepted into the BlueDot Impact AI Safety program. Having spent years building systems that work, I am now dedicated to ensuring that the most powerful systems we ever create—AGI—are safe and aligned with human values.

Beyond the Code

When I’m not auditing neural networks or debugging cloud infrastructure, you’ll find me: