iiinova — where your digital self keeps evolving

We use weight pruning here as a causal probe of model internals, not a deployment-ready defense. But — this opens a path toward *mechanistic alignment*: considering the mechanisms behind harmful behavior, rather than training behavioral guardrails on top of them.

And it's specifically the generative mechanism we remove: fine-tuning on harmful examples can mostly restore it—confirming our point that the underlying knowledge is largely intact.

100

Generating harmful content is dissociated from "understanding" it. Pruned models retain nearly full ability to detect, explain, and refuse harmful requests. Refusal and generation are *double dissociated*: pruning one leaves the other intact.

And here as well, it generalizes between domains (plus, we observe significant intersection of weight sets).

This also explains emergent misalignment ( @BetleyJan et al.) If harmful behaviors share weights, then fine-tuning one narrow domain can unintentionally affect others. → Pruning a narrow misaligned domain substantially reduces emergent misalignment.

This compression seems to be a product of alignment training. Aligned models show far greater separation between harmful and benign weights than their pretrained counterparts. Alignment reshapes the internals, even when behavioral guardrails remain brittle. [more in the paper]

The mechanism is unified across harm types. E.g., prune weights identified from malware generation → hate speech drops too. Different harms rely on a shared underlying mechanism. These weights also heavily overlap.

120

Paper >> arxiv.org/abs/2604.09544 We use weight pruning to probe model internals. Result: pruning ~0.0005% of model parameters, harmful generation drops dramatically, while general capabilities remain largely intact.

New paper: LLMs encode harmful content generation in a distinct, unified mechanism Using weight pruning, we find that harmful generation depends on a tiny subset of the weights that are shared across harm types and separate from benign capabilities. 🧵

172

actionable-interpretability-guide.github.io

Full paper >> actionable-interpretability-guide.github.io/paper.pdf Blog >> actionable-interpretability-guide.github.io

000

We’re not saying all interpretability work must be immediately actionable— curiosity-driven research still matters. But actionability is a high bar: understanding that works outside the lab. To make your next project more actionable, use our checklist >>

120

Actionable interpretability is worth aiming for. We identified five domains where answering *why* unlocks a fundamental advantage.

100

Interpretability isn't actionable (yet) for three reasons: → Papers aren't expected to demonstrate applications → Insights are shown in oversimplified settings without real baselines → Methods require domain expertise

100

Why haven't insights from interpretability transformed AI yet? Because we're not prioritizing actionable insights. Full paper >> actionable-interpretability-guide.github.io/paper.pdf Blog - The Hitchhiker's Guide 🧭 to Actionable Interpretability >> actionable-interpretability-guide.github.io/

Hadas Orgad @hadasorgad.bsky.social · 03/05/2025

Our ICML 2025 workshop on Actionable Interpretability drew massive interest. But the same questions kept coming up: What does "actionable" mean? Is it achievable? How? We're ready to answer. 🧵

12410

Deadline extended! ⏳ The Actionable Interpretability Workshop at #ICML2025 has moved its submission deadline to May 19th. More time to submit your work 🔍🧠✨ Don’t miss out!

043

Reposted by Hadas Orgad

Aaron Mueller @amuuueller.bsky.social · 23/04/2025

Lots of progress in mech interp (MI) lately! But how can we measure when new mech interp methods yield real improvements over prior work? We propose 😎 𝗠𝗜𝗕: a 𝗠echanistic 𝗜nterpretability 𝗕enchmark!

15115

Hadas Orgad @hadasorgad.bsky.social · 31/03/2025

• Model Innovation – Designs and training inspired by interpretability. • Impact Measurement – Benchmarks for real-world effectiveness. • Critical Perspectives – Feasibility, limits, and future directions. Website >>> actionable-interpretability.github.io

actionable-interpretability.github.io

General Information

ICML 2025 - Vancouver

030

Hadas Orgad @hadasorgad.bsky.social · 31/03/2025

• Real-world Applications – Tackling bias, hallucinations, adversarial threats, and use in critical domains like healthcare, finance and cybersecurity. • Method Comparison – Interpretability vs. alternative methods such as fine-tuning, prompting, etc.

120

Hadas Orgad @hadasorgad.bsky.social · 31/03/2025

We aim to foster discussions on how interpretability research can inform concrete improvements in model design, safety, and robustness. Topics of interest: ⬇️