🔬

Studying Large Language Model Generalization with Influence Functions

📅 2023
technical research breakthrough
🔵 Rare

📖 Description

Studying Large Language Model Generalization with Influence Functions Roger Grosse?:, Juhan Bae?:, Cem Anil?: Nelson Elhage; Alex Tamkin, Amirhossein Tajdini, Benoit Steiner, Dustin Li, Esin Durmus, Ethan Perez, Evan Hubinger, Kamil? e Luko?i? ut? e, Karina Nguyen, Nicholas Joseph, Sam McCandlish Jared Kaplan, Samuel R. Bowman Abstract When trying to gain better visibility into a machine learning model in order to understand and mitigate the associated risks, a potentially valuable source of evidence is: which training examples most contribute to a given behavior? Influence functions aim to answer a counterfactual: how would the model?s parameters (and hence its outputs) change if a given sequence were added to the training set? While influence functions have produced insights for small models, they are difficult to scale to large language models (LLMs) due to the difficulty of computing an inverse-Hessian-vector product (IHVP). We use the Eigenvalue-corrected Kronecker-Factored App...

📊 Game Impacts

Not verified in game

Which variables this event was proposed to move, and in which direction. The magnitudes are held in the corpus but are not shown here, because they have not been verified against the shipped game. They come from pdoom-data. They describe what an event was proposed to do, not what the shipped game does with it. Most events in the corpus are flavour: they are shown for colour and do not move any game variable. Only a small minority reach the systems below, and several of the variables listed here are not read by the game at all yet. Treat this table as a design proposal under review, not as a measurement of play. Corrections and arguments are welcome — the suggestion links at the foot of this page go straight to the data repo.

Variable Direction Condition
Research proposed: up Always
Papers proposed: up Always
Vibey Doom proposed: up Always

💭 Reactions

🔬 Safety Researcher Reaction: ⚠️ Placeholder - Needs Real Quote
"Important contribution to the field"
📰 Media Reaction: ⚠️ Placeholder - Needs Real Quote
"Academic research release"
💡 Found a Real Quote? Suggest it here

🔗 Sources

🏷️ Event Metadata

Think this event's metadata could be improved? Category and tags describe the real-world event and are maintained in pdoom-data. Rarity, game impacts and p(doom) effects are game-mechanical values owned by pdoom1. Each link below goes to the repository that decides that field.

📁 Category Technical Research Breakthrough → Suggest different category
⭐ Rarity 🔵 Rare → Suggest different rarity
🏷️ Tags (0) → Suggest tag changes
📊 Game Impacts (3) 3 variable changes → Suggest impact changes
☢️ p(Doom) Impact No direct impact → Suggest p(doom) change
📝 General Metadata Year, description, reactions → Comprehensive review

🤝 Found an Issue?

This event data is sourced from the pdoom-data repository. If you notice errors or want to suggest improvements:

GitHub Issue (Preferred) 📧 Email (No GitHub)
← Back to All Events