Latent Thought
What is Latent Thought?
Latent Thought is a blog where I share my research and ideas across mathematics, AI, psychology, and philosophy.
Work
DAN prompts for SOTA LLMS
The paper shows how a simple jailbreak prompt can bypass guardrails in top LLMs with 92% success, using role-play, message injection, and obfuscation
Scaling TTC
AI models can reason better by combining Chain-of-Thought prompting with Reinforcement Learning, but they still struggle with real-world tasks and limited training data. New methods like self-play and synthetic data generation offer promising ways to improve them.
LLM Search Trees
A search based approach to LLM's akin to alpha go
Mind Palace Blog
Exploration of novel use cases for mind palace
Self-stop Layer: Neural Network Architecture Improvement Proposal
Neural network architecture designed to emulate the brain's efficiency by reutilizing its neurons multiple times to perform a single computation.
Overfitting Adjust Regresion
A paper exploring a new and more accurate form of regression inspired by kernel regression
Why did I host my research on here?
01
0 limitations
I want to publish papers not beholden to the arbitrary criteria of research committees, or convoluting incentives of academia. Lot's of my research is closer to blogger nature. However, most of my papers are also hosted on arxiv.
03
0 bullshit
Unlike the majority of academic work, I hope to release work that can be understood and appreciated by any reader.
02
0 restrictions
As an undergrad researcher, I'm beholden to the direction of professors. On Latenet Thought I'm able to explore whatever I want however I want including art, poetry, polotics and philosophy.
04
0 wait
Work often takes years to publish, but on my own blog there is 0 wait, so I can release work that is relevant, topical and up to date.