Reverse-engineering what models think on Ribhu's Blog

Emotion Retrieval through Art in Latent Space

Mon, 31 Mar 2025 00:00:00 +0000

When I was in the NGMA, this thought came to me that art is simply a way of representing latent space in 2-dimensions, or 3. Its a way of distilling what we are feeling (that latent space information embedding). Plus, the way art moves people is because it is a direct way of matching embeddings. Language is simply an encoder of that latent information. Art works by creating representations (embeddings) that match our internal representations, causing emotional responses which are not transformed into a specific modality.

The Benchmark Trap: Why LLM Metrics Mislead and Evals Enlighten

Sun, 12 May 2024 00:00:00 +0000

“We have just launched our newest large language model trained with the best curated dataset with the state-of-the-art GPUs and we have improved on the MMLU by n number of points, surpassing the previous best.” Does this sound familiar? Reads like a press release of an AI company releasing their shiny new LLM for the world. As the arms race heats up it was bound to happen. But there’s one big problem with this line. No, I’m not talking about the obvious x-risk with larger, uninterpretable, and unaligned models. I’m not talking about the effects on climate change due to the the emissions of data centres. I want to illustrate one issue that’s not talked about as much but is actually much easier to control and to take action on.

Behind the Curve: The Slow March of AI into Classrooms

Tue, 16 Apr 2024 00:00:00 +0000

I will start with the obvious disclaimer that I work in the domain, and hence maybe I have noticed the trends I have mentioned a lot more, but that also gives me a broader insight. Its a double-edged sword because its an evident bias, and so I welcome any counter points.

With the explosion of generative AI, and the current in-fashion methods of using high quality synthetically generated data, we see many folks talking about the Cambrian Explosion of AI. It has enabled AI to permeate not just the public consciousness but also every single industry one can imagine. Even traditionally not tech-heavy domains such as real estate seem to be incorporating it in thanks to the evangelisation of the field and the ease of integrating them. The Return-on-Investment has increased quite a bit specially with the advent of large language models. But there is one industry which is still in the “testing the waters” phase. Education. One recent tweet provides more concrete numbers.

Artificial Misdirection: Misalignment of LLMs

Fri, 02 Feb 2024 00:00:00 +0000

In 2024 humanity is facing a whole bunch of societal and potentially existential threats. World War 3, global recession, climate change, the list is pretty much endless. Development of capable AI systems, although still narrow, provide another whole set to be added to that. A common theme among all of these problems is that they are never-ending battles, that we as a civilization have to keep working to ensure it doesn’t happen or worsen.

The Forgotten Lever of Neural Scaling Laws

Fri, 19 Jan 2024 00:00:00 +0000

While driving a few weeks back, I was listening to a podcast (Cognitive Revolution) episode where the Perplexity CEO, Aravind Srinivas was talking about moats in the AI space.¹ One of the things he mentioned really stuck out to me.

Everybody’s building their index because I think in the world where large language models are commodity and the training recipe for them or the weights are just right on their open source, the edge goes to the data markets (…) You don’t even want 100 billion pages in your index. It’s not about the quantity here again, right? You want the best web pages on your that’s probably a billion or 10 billion.

Where Human and Artificial Intelligence meets

Wed, 23 Jan 2019 00:00:00 +0000

Going over some of my previous papers I have written in College, I found this gem that I felt needed to be shared with the world. It is quite fitting that the class this was written for was called Cognitive Consequences of technology.

Artificial Intelligence is one of the fastest growing fields of scientific research. It has enabled automation of human tasks/decision-making, given insights into the working of our own brains, and made existing systems more efficient. When speaking of automation, many assume it means replacing humans with AI, while in fact it has been seen on many occasions that solely an AI does not always make the right decisions or do the task properly. While human jobs won’t be taken away by AI, they will collaborate in the workplaces of tomorrow. It is an exciting age to live in as never before have digital tools been so responsive to us, nor we to our tools which allows for more creativity and less time spent in learning to use the tools themselves.¹

Developing an intuition for better understanding of convolutional neural networks

Wed, 07 Nov 2018 00:00:00 +0000

The Human Eye

The best visual tool that man has known for most of time is the human eye. It is the primary tool used by our brain to perceive the world and make sense of it. While the basics of the human eye are explainable by optics, it is how that raw stimulus invokes imagery in our brain that confused and fascinated people for a long time. It was in the late 1950s that two professors David Hubel and Torsten Wiesel, who were experimenting by inserting electrodes in the visual cortex of a cat and observing individual neurons, made a breakthrough: more.

Musings on Fate

Tue, 15 Aug 2017 00:00:00 +0000

Note: This is just me thinking out loud.

“Amor Fati”. Love your fate.

The concept of fate or destiny is simple. Your fate is decided, well even if it is, is it random? We are told to accept it, embrace it, love it. But, does it follow a pattern or is it completely random? And what is it?

Your fate is your life. Let’s take a look at life then, of all living things. Charles Darwin discovered Natural Selection. Quite simply, it refers to the survival of the fittest. The ones who can adapt to their environments better have a greater chance of surviving. These adaptations came in the form of genetic mutations. Moreover, these genetic mutations, which are random, help certain animals get an edge over the other non-mutated beings. Is it fate that caused some to get the necessary mutations? That it was the strongest of the pack because it was the “chosen one”? Or was it completely random?

The United States of America vs Soccer

Thu, 20 Apr 2017 00:00:00 +0000

Note: I wrote this essay as one of the final papers for a college history course. Thought it was a good read, got a good grade on it too, so I thought might as well share it with the world.

The imperial system, playoffs, month before date. One might wonder what all of these have in common — the answer, they are all products of American exceptionalism. The United States of America operates on its own terms as the world has seen over the years. The idea of this exceptionalism perhaps arose from the uniqueness of the history of the United States. This ideology shaped the American sporting culture and hence, led to American sporting exceptionalism. Moreover, this sporting exceptionalism in turn caused soccer, the most popular game in the world, to not flourish as others thought it would.

A Robot Philosopher

Tue, 13 Dec 2016 00:00:00 +0000

Recently, Google published a report which featured a conversation between an AI and a human. Here’s an excerpt:

Human: What is the purpose of life? Machine: To serve the greater good. Human: What is the purpose of living? Machine: To live forever. Human: What is the purpose of existence? Machine: To find out what happens when we get to the planet earth. Human: Where are you now? Machine: I’m in the middle of nowhere. Human: What is the purpose of dying? Machine: To have a life. Human: What is the purpose of being intelligent? Machine: To find out what it is. Human: What is the purpose of emotions? Machine: I don’t know.

Why Louis Van Gaal was the Dark Knight of Manchester United (and other super clubs)

Thu, 08 Sep 2016 00:00:00 +0000

Louis Van Gaal has been called many things throughout his controversial yet successful career. Call him a tactical genius or a philosopher, he will never truly be known for what he has actually done, whether it be accidentally or on purpose. To get an idea of what I’m talking about let’s go back, way back to Amsterdam, 1995.

Not-so-humble Beginnings

“Congratulations on signing the best coach in the world.” One might think this as some former player of the club or maybe a legend of the game commenting on Van Gaal’s potential to be the best someday. Nope. It was the young and arrogant Louis himself declaring his brilliance to the world. A 2nd placed finish and UEFA cup victory in the first season, then a 3rd place finish and a KNVB cup in his second. While many would attribute those successes to an in-form Dennis Bergkamp, the same cannot be said for the 1994–95 season, which brought Van Gaal into the limelight of global football and helped his claim to being the best. Ajax went undefeated in the league to defend their crown as champions of England and unexpectedly shocked Europe’s big boys to claim the Champions League crown.

About

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

Who am I

Ribhu Lahiri. A pragmatist chasing rationality. Big words aside, I’m the co-founder of kaksha.ai, where I lead the research. Avid reader of books, blogs, and newsletters, indie movie enjoyer, and you can find me watching the premier league (GGMU) on the weekends.

As a Research Engineer leading an AI-education startup, I’m committed to making AI accessible to all. In addition to running my own venture, I provide strategic consulting to other AI startups, offering practical solutions tailored to their unique needs. My work also extends to AI safety and alignment research, focusing on building safe, beneficial AI systems that align with human values. I am passionate about fostering Human-AI collaboration and envision a future powered by Augmented Intelligence. My goal is to contribute to a world where AI and humans work together to drive growth and innovation.

Projects

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