Lossfunk Letters

Your LLM is a confused oracle

We show that the forecasting accuracy of LLMs depends on what you ask and how you ask

Nov 26 • Chinmay and Paras Chopra

7

2

3

Future of LLMs might not be Autoregressive

Intro to the world of block diffusion

Nov 24 • Ayush Nangia and Aman Gokrani

10

4

Sequential scaling outperforms parallel scaling for LLMs

AI reasoning just got a upgrade: At the same compute cost, sequential thinking—iteratively refining ideas—beats parallel "crowdsourcing" in 95.6% of…

Nov 6 • Aman Sharma

5

3

October 2025

Notes on Tiny Recursion Network

aka how a 7M parameter network gets SOTA on Sudoku-Extreme with 87% accuracy

Oct 31 • Paras Chopra

4

1

Do LLMs know when they've gotten a correct answer?

We show they do and then use it to help cut reasoning cost (in tokens) by up to 50% without losing accuracy

Oct 29 • Aman Sharma

14

4

How do LLMs "think" across languages

Performance of LLMs differ based on language on reasoning tasks and this difference varies for each task.

Oct 28 • Shourya

14

5

3

What's the point of doing research?

The fun of the struggle is the point

Oct 17 • Paras Chopra

35

4

7

September 2025

How to choose research problems

TLDR: balance between what your heart says and what the community will value

Sep 10 • Paras Chopra

10

1

3

Don't tell LLMs you are an AI safety researcher

Adding “for AI safety research” increased refusals on a harmless paraphrasing task for some top models. Conservative, aligned public models might be…

Sep 5 • Dhruv Trehan

7

4

3

August 2025

Tips on writing your first research paper

aka, how to increase your odds of acceptance at an AI/ML conference

Aug 29 • Paras Chopra

10

1

What is research and how to do it?

or, can you teach an AI how to do good research?

Aug 12 • Paras Chopra

61

7

12

Notes on Hierarchal Reasoning Model

aka what can we learn from the brain!

Aug 6 • Paras Chopra

9

1

1