Hey there, welcome to my blog! The purpose of this space is to share my ongoing research endeavours, occasional thoughts and to provide in-depth views into my previous work which can help others working on the same problems.

Following Hamming’s “You and Your Research”, I think one of the most critical problems in this field is developing an ability for causal discovery in language models with oversight to minimise AI X-Risk and better understand the nature of reality.

This blog traces my footprints on this direction. Feel free to contact me on https://x.com/Diksha1713 or drop a mail at [email protected].

Cheers!

https://diksha-shrivastava13.github.io/

****The Causal Discovery Series | Diaries | Adjacent Curiosities

[Exciting Research Directions](<https://diksha-shrivastava13.notion.site/Exciting-Research-Directions-1fee9b1b1b2a8090bf93f3920619c89b>) | [Product Experience](<https://diksha-shrivastava13.notion.site/Product-Experience-202e9b1b1b2a806cbd9ac484c670a1ba>) 

https://diksha-shrivastava13.github.io/

All views are my own. Some of the work done here had sensitive data which is replaced with

analogies to make the content easier to understand. For any issues, please mail me.

The Causal Discovery Series

1. Stumbling upon a problem while developing a high-stakes product
2. The Investigation Begins
3. Can creative engineering hacks overcome this?
4. Roadblock is a fundamental capability issue!
5. What are the implications?
6. Research Statement 0.1
7. Causality, ML and Physics—What is Causal Discovery?
8. From Capability to X-Risk
9. Research Statement 1.0
10. What comes next after Scientist AI?
11. Is the universe truly a mathematical structure?
12. Research Statement 1.1

⚠️ Work In Progress

Exciting Research Directions

1. Causal Discovery in Evolving Curricula for Causal Understanding with Human Oversight
2. Externalising Latent Reasoning to an Interpretable Dynamic World Model
3. The Problem of Perspectives I: Studying the Effects of Perspective on Causal World Models for Alignment
4. The Problem of Perspectives II: Making AI Ask the Right Questions in a Scientific Dialect
5. Study of Causality, Time and Perspective
6. Formal Verification for Scientific Reasoning

🎬 Where it Began —

Can language models formulate ML problems from deep, interacting subsystems?

⚠️ Work In Progress

This Blog is curated like a garden—quietly grown, occasionally overgrown, always alive.

Adjacent Curiosities

1. Modeling Wayfinding: A Hybrid Neurosymbolic and RL Approach to Dynamic Decision-Making in Quest-Driven Narrative Worlds
2. AI for Astrophysics: Automating Domain Specific Tasks with Large Language Models
3. Complete Project Report: 50+ Iterations of Agentic Reasoning in a Hierarchical, Complex World Model for Country Policy Decisions
4. Agents are Decision-Makers First: Leveraging Graph of Decisions for Intermediate Reward Modeling
5. Reasoning Beyond Correctness: Problem Generation through Divergent Rearrangement Sampling [Paper In Review]
6. Closing the Loop: Execution-Guided Continuous Generation for Adaptive Model Reasoning
7. ChessGAN: Making an Agent Lose Chess Games Exactly 50% of the time & Thoughts on the Turing Test
8. Symbolic Music Generation: Can Music be Designed to Help Neurodivergent People?
9. First Research Endeavour: Analysis of Neural Correlates of Different Music Genres using Machine Learning
10. Ancient History: The First Ever Startup Idea on Immersive Generative Environments for Therapy and Turing Test which got me into AI.

Product Experience

Believe it or not, I’ve shipped 3 products. Research is a natural progression.

1. [For SAP] How can a LLM-based system continuously learn from feedback to refine its retrieval and reasoning?

2. [For Federal Ministry BMZ] Can AI reason across multi-subsystem policy decisions spanning decades?

   — Not for long and not reliably without causal discovery.

3. [Stealth Start-up] Can language models formulate ML problems from a high-level description and automate ML cycles with abstract data?

I love the thrill of shipping every week and investor presentations…in hindsight, it’s a guilty pleasure. I’ve met the best people this way, and might go back some time.

Diaries

Felt the need to leave a list of fun facts about me in this blog:

• I’m an avid reader and have written two science-fiction novellas named “To Eternity and Beyond” and “A Walk In The Rain” along with a handful of short stories.

• I think the ability to imagine is a priceless gift. The best thing my parents did was to surround me with books as a kid. The highest praise I’ve received on my writing was a special invite from the Government of India.

• I used to organise Asia’s writing sprints for National Novel Writing Month (NaNoWriMo) in early teens.

• I’ve had an environmentalist streak throughout my teenage, which meant leading National Children’s Science Congress teams and conducting studies on climate change, pollution, birds and the like.

• Later I understood that change can be best made top-down, around the same time I fell headfirst for calculus. I can’t explain why, though reading ‘A Brief History of Time’ perhaps contributed to it.

• Once upon a time, I had five dogs. Growing up in a college campus meant friendliness with most animals, though all of them have a tendency to slap me.

• For the life of me, I can’t dance or socialise or <insert any extroverted activity> unless very specific conditions are met. I’m perfectly comfortable just stringing along.

• I love, love, love mountains and rainy weather and walking and spicy food. I can play classical music.

• I’d consider life complete if I can just run a small research lab on AI for Discovery and spend my time peacefully researching the problems which seem the most important to me.

  Personal Notes

  I like volunteering and open-source, having volunteered for 9 technical societies so far, however my dedication depends on the dedication around. Time is Precious.

Putting together this timeline from the first semester of my bachelors to the last was a sad thing indeed, especially considering how few of these end up truly meaning something. This was without considering the long list of side quests I learnt ML from…

Maybe someday I’ll get around to filling all the gaps in between.

Mischief Managed!