Papers › Beyond Toxic Neurons: A Mechanistic Analysis of DPO for Toxicity Reduction

Beyond Toxic Neurons: A Mechanistic Analysis of DPO for Toxicity Reduction

10 Nov 2024arXiv:2411.06424archive 2025-07-28

Yushi Yang, Filip Sondej, Harry Mayne, Adam Mahdi

Safety fine-tuning algorithms are widely used to reduce harmful outputs in language models, but how they achieve this remain unclear. Studying the Direct Preference Optimization (DPO) algorithm for toxicity reduction, current explanations claim that DPO achieves this by dampening the activations of toxic MLP neurons. However, through activation patching, we show that this explanation is incomplete. Projections onto a toxicity probe's direction show that only 4.9% of toxicity reduction comes from dampened toxic neurons. Instead, DPO reduces toxicity through distributed activation shifts across a majority of neurons, progressively shifting MLP layer outputs away from toxicity. These shifts accumulate across four neuron groups: two reducing toxicity and two promoting anti-toxicity. Activation patching validates the cumulative roles of these groups, where patching all identified groups effectively replicates DPO's effects. These findings illustrate DPO's mechanism: it reduces toxicity by accumulating small activation shifts across many neurons throughout the layers. Our findings provide new mechanistic insights into how safety fine-tuning reduces harmful outputs in language models.

Code

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

15 samples harvested; 0 ran; 0 honoured the contract we drafted; 15 have no recorded run. Read from Syntology's graph 2026-09-24; that is when this build read the record, not when the samples ran.

Licence: 0 of the 15 samples are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from yushi-y/dpo-toxic-neurons. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself; “our draft was wrong” and “fixture could not drive it” are failures of Syntology's instrument, not of the code.

Each sample ends with its code_sha256, Syntology's identity for that exact code. An agent fetches the stored sample with Syntology's MCP tool get_code(code_sha256="…") (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted. community: Not in the archive's code links for this paper; a community repository Syntology harvested. Samples from a repository marked official are listed first. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of four reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; or the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar). Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections