whatisgithub

What is lastwordwinscot?

gpgabriel25/lastwordwinscot — explained in plain English

Analysis updated 2026-05-18

0PythonAudience · researcherComplexity · 4/5LicenseSetup · hard

In one sentence

A research paper and reproducible codebase showing that many chain-of-thought corruption studies accidentally measure where a model's answer text sits, not which reasoning steps actually matter.

Mindmap

mindmap
  root((LastWordWinsCoT))
    What it does
      Studies CoT faithfulness
      Finds a format confound
      Tests five model families
    Tech stack
      Python
      JAX
      Cloud TPU
    Use cases
      Reproduce paper figures
      Study experiment design
      Rerun TPU inference
    Audience
      AI researchers
      Students
    Contents
      Paper and figures
      Eleven datasets
      81 result files

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

What do people build with it?

USE CASE 1

Reproduce the paper's figures and statistics directly from the included datasets and result JSONs without needing a TPU.

USE CASE 2

Study the dataset-building and analysis scripts as a template for running your own chain-of-thought faithfulness experiments.

USE CASE 3

Rerun the original model inference experiments on a Cloud TPU to verify or extend the findings.

What is it built with?

PythonJAXTPU

How does it compare?

gpgabriel25/lastwordwinscot0xhassaan/nn-from-scratch3ks/embedoc
Stars00
LanguagePythonPythonPython
Last pushed2023-06-08
MaintenanceDormant
Setup difficultyhardmoderatehard
Complexity4/54/51/5
Audienceresearcherdeveloperdeveloper

Figures from each repo's GitHub metadata at analysis time.

How do you get it running?

Difficulty · hard Time to first run · 1day+

Full reproduction of model inference needs a Cloud TPU v5e and JAX, analysis and figures alone only need the included data.

Code is MIT, free to reuse for any purpose. Paper, data, and figures use Creative Commons Attribution, reusable including commercially if you credit the author.

So what is it?

This repository holds the paper, code, datasets, and result files for a research project about how AI models explain their reasoning, known as chain of thought, or CoT. The author set out to study how much of that reasoning is genuine versus after the fact justification, and along the way found a bigger problem with how researchers test this in the first place. The main finding is that many studies which corrupt a step in a model's reasoning to see if accuracy drops are partly measuring the wrong thing. Most benchmark reasoning chains end with a line like "the answer is X". If you damage the text near the end of the chain, you are often just damaging the spot where the answer sits, not the part of the reasoning doing real work. Move that answer line elsewhere, and the part that looks important moves with it. The paper runs several experiments to test this, including comparing five different model families, and finds the effect holds up under repeated attempts to disprove it. The repo is organized so others can reproduce the work. It includes the LaTeX source and PDF of the paper, the figures, eleven curated datasets in JSONL format, eighty one result files with raw numbers, and Python scripts split into dataset building, model inference, analysis, and figure generation. Running the analysis and figure scripts only needs the included data. Rerunning the original model inference requires a Cloud TPU and the JAX machine learning framework, since that is what the experiments were run on. This is aimed at researchers and students interested in how trustworthy AI reasoning explanations really are, not at general software users. The code itself uses MIT licensing, while the paper, figures, datasets, and results use a Creative Commons Attribution license that allows reuse and commercial use as long as the author is credited.

Copy-paste prompts

Prompt 1
Explain in plain terms what a format confound is in the context of chain-of-thought corruption studies, using this repo's findings as an example.
Prompt 2
Walk me through how to reproduce the GSM8K format ablation figure from the JSONs in results/summary without running any model inference.
Prompt 3
Summarize the three-step protocol this paper proposes for future chain-of-thought faithfulness research.
Prompt 4
What would I need to change in the inference scripts to run this experiment on a different model family?

Frequently asked questions

What is lastwordwinscot?

A research paper and reproducible codebase showing that many chain-of-thought corruption studies accidentally measure where a model's answer text sits, not which reasoning steps actually matter.

What language is lastwordwinscot written in?

Mainly Python. The stack also includes Python, JAX, TPU.

What license does lastwordwinscot use?

Code is MIT, free to reuse for any purpose. Paper, data, and figures use Creative Commons Attribution, reusable including commercially if you credit the author.

How hard is lastwordwinscot to set up?

Setup difficulty is rated hard, with roughly 1day+ to a first successful run.

Who is lastwordwinscot for?

Mainly researcher.

Open on GitHub → Ask about another repo

This repo across BitVibe Labs

Verify against the repo before relying on details.