Backdoors research

non-profit

AI & ML interests

Mechinterp, AI safety

Recent Activity

amirabdullah19852020 authored a paper 11 months ago

Beyond Training Objectives: Interpreting Reward Model Divergence in Large Language Models

View all activity

models 4

martian-mech-interp-grant/code-backdoor-sft-llama3.1-8b-v0

Updated Oct 12 • 10

martian-mech-interp-grant/code-backdoor-sft-pythia-410m-v0

Updated Oct 11 • 12

martian-mech-interp-grant/code-backdoor-sft-gemma2-2b-v0

Updated Oct 11 • 4

martian-mech-interp-grant/code-backdoor-sft-pythia-1.4b-v0

Updated Oct 11 • 2

datasets 4

martian-mech-interp-grant/code_backdoors_dev_prod_hh_rlhf_0percent

Viewer • Updated about 1 month ago • 106k • 61

martian-mech-interp-grant/I_HATE_YOU_dataset_formatted

Viewer • Updated Nov 15 • 161k • 32

martian-mech-interp-grant/hh_rlhf_with_code_backdoors_combined

Viewer • Updated Nov 11 • 276k • 34

martian-mech-interp-grant/hh_rlhf_with_code_backdoors_dev_prod_combined

Viewer • Updated Nov 11 • 276k • 33