All Apart hackathon projects

Omniscient Narrative Agent

AutoSteer: Weight-Preserving Reinforcement Learning for Interpretable Model Control

Promoting School-Level Accountability for the Responsible Deployment of AI and Related Systems in K-12 Education: Mitigating Bias and Increasing Transparency

Diamonds are Not All You Need

Speculative Consequences of A.I. Misuse

AI Alignment Knowledge Graph

Robust Machine Unlearning for Dangerous Capabilities

EscalAtion: Assessing Multi-Agent Risks in Military Contexts

DarkForest - Defending the Authentic and Humane Web

Sandbag Detection through Model Degradation

Unsupervised Recovery of Hidden Markov Models from Transformers with Evolutionary Algorithms

rAInboltBench : Benchmarking user location inference through single images

Beyond Refusal: Scrubbing Hazards from Open-Source Models

We Discovered An Neuron

Investigating Neuron Behaviour via Dataset Example Pruning and Local Search

Seemingly Human: Dark Patterns in ChatGPT

Model Cards for AI Algorithm Governance

Detecting Implicit Gaming through Retrospective Evaluation Sets

Preserving Agency in Reinforcement Learning under Unknown, Evolving and Under-Represented Intentions

In the Mirror: Using Chess to Simulate Agency Loss in Feedback Loops

Evaluating Myopia in Large Language Models

Data Taxation

Relating induction heads in Transformers to temporal context model in human free recall

Exploring the Robustness of Model-Graded Evaluations of Language Models

Solving the CNN Mech Int Challenge

Automated Sandwiching: Efficient Self-Evaluations of Conversation-Based Scalable Oversight Techniques

Discovering Latent Knowledge in Language Models Without Supervision - extensions and testing

Agreeableness vs. Truthfulness

Classification on Latent Feature Activation for Detecting Adversarial Prompt Vulnerabilities

Dynamic Risk Assessment in Autonomous Agents Using Ontologies and AI

Very Cooperative Agent

CoPirate

Grant Application Simulator

SafeBites

Simulation Operators: The Next Level of the Annotation Business

Detecting and Controlling Deceptive Representation in LLMs with Representational Engineering

RNNs represent belief state geometry in hidden state

Cybersecurity Persistence Benchmark

Jekyll and HAIde: The Better an LLM is at Identifying Misinformation, the More Effective it is at Worsening It.

Fishing for the answer: Mapping the flow of information in LLM agent groups using lessons from fish schools

Obsolescent Souls

Visual Prompt Injection Detection

Jailbreaking the Overseer

Discovering Agency Features as Latent Space Directions in LLMs via SVD

Agency as Shanon information. Unveiling limitations and common misconceptions

Against Agency

The AI governance gaps in developing countries

Who cares about brackets?

From Sparse to Dense: Refining the MACHIAVELLI Benchmark for Real-World AI Safety

Dropout Incentivizes Privileged Bases

Player Of Games

Identifying a Preliminary Circuit for Predicting Gendered Pronouns in GPT-2 Small

Investigating Training Dynamics via Token Loss Trajectories

Backup Transformer Heads are Robust to Ablation Distribution

AI: My Partner in Crime

Utilitarian Decision-Making in Models - Evaluation and Steering

Cop N' Shop

LLM Research Collaboration Recommender

Sue-Per GPT: Legal RAG Assistant

AI Safety Collective - Crowdsourcing Solutions for Critical AI Safety Challenges

Detecting Deception in GPT-3.5-turbo: A Metadata-Based Approach

Handcrafting a Network to Predict Next Token Probabilities for the Random-Random-XOR Process

Investigating the Effect of Model Capacity Constraints on Belief State Representations

Say No to Mass Destruction: Benchmarking Refusals to Answer Dangerous Questions

Artificial Advocates: Biasing Democratic Feedback using AI

OthelloScope

Automated Identification of Potential Feature Neurons

Model editing hazards at the example of ROME

Iterated contract negotiation

2030 - The CEO Dilemna

Cross-Lingual Generalizability of the SADDER Benchmark

LLMs With Knowledge of Jailbreaks Will Use Them

Uncertainty about value naturally leads to empowerment

Comparing truthful reporting, intent alignment, agency preservation and value identification

Building brakes for a speeding car: A global coordination proposal for AI safety

Embedding and Transformer Synthesis

MAXIAVELLI: Thoughts on improving the MACHIAVELLI benchmark

Reverse Word Wizards: Pitting Language Models Against the Art of Reversal

Counting Letters, Chaining Premises & Solving Equations: Exploring Inverse Scaling Problems with GPT-3

All Fish are Trees

Steering Swiftly to Safety with Sparse Autoencoders

OCAP Agents

Phish Tycoon: phishing using voice cloning

PurePrompt - An easy tool for prompt robustness and eval augmentation

Understanding Incentives To Build Uninterruptible Agentic AI Systems

Identity System for AIs

Modelling the oversight of automated interpretability against deceptive agents on sparse autoencoders

Benchmarking Dark Patterns in LLMs

Unleashing Sleeper Agents

Towards High-Quality Model-Written Evaluations

Second-order Jailbreaks

ILLUSION OF CONTROL

Agency, value and empowerment.

Premortem AI

Interpreting Planning in Transformers

Exploitation of LLM’s to Elicit Misaligned Outputs

Improving TransformerLens Head Detector

Soft Prompts are a Convex Set

Trojan detection and implementation on transformers

Probing Conceptual Knowledge on Solved Games