All Apart hackathon projects

Reasoning with Chain of Thought

Wording influences truthfulness

Othello Mechint playground

Detecting Phase Transitions

Simulating an Alien

Exploring OthelloGPT

User Transparency Within AI

Community-First: A Rights-Based Framework for AI Governance in India's Welfare Systems

National Data Privacy and Governance Act

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

Implementing a Human-centered AI Assessment Framework (HAAF) for Equitable AI Development

A Critical Review of "Chips for Peace": Lessons from "Atoms for Peace"

AI Monitoring as a Rapid and Scalable Policy Solution: Weekly Global Bulletins on AI Developments

Grandfather Paradox in AI – Bias Mitigation & Ethical AI1

Glia for Healthcare Organisations

A Fundamental Rethinking to AI Evaluations: Establishing a Constitution-Based Framework

Hero Journey: Personalized Health Interventions for the Incarcerated

mHeatlth Ai

Glia

AI Parliament

Policy Analysis: AI and Sustainability: Climate Impact Monitoring

applai

Digital Rebellion: Analyzing misaligned AI agent cooperation for virtual labor strikes

Next-Gen AI-Enhanced Epidemic Intelligence

AI and Public Health: TSA Pre Health Check

Mapping Intent: Documenting Policy Adherence with Ontology Extraction

EcoNavix

Towards a Unified Framework for Cybersecurity and AI Safety: Recommendations for Secure Development of Large Language Models

Enviro - A Comprehensive Environmental Solution Using Policy and Technology

Enhancing Human Verification Systems to Address AI Agent Circumvention and Attributability Concerns

Politicians on AI Safety

Policy Framework for Sustainable AI: Repurposing Waste Heat from Data Centers in the USA

Predictive Analytics & Imagery for Environmental Monitoring

Proposal for U.S.-China Technical Cooperation on AI Safety

Proposal for a Provisional FDA Designation Targeting Biomedical Products Evaluated with Novel Methodologies

Infectious Disease Outbreak Prediction and Dashboard

Cross-model surveillance for emails handling

Inference-Time Agent Security

Intent Inspector - Protecting Against Prompt Injections for Agent Tool Misuse

AI Honeypot

AI Agent Capabilities Evolution

An Autonomous Agent for Model Attribution

Using ARC-AGI puzzles as CAPTCHa task

LLM Agent Security: Jailbreaking Vulnerabilities and Mitigation Strategies

AdGPT

Interpreting a toy model for finding the maximum element in a list

Finding Circular Features in Gemma 2 2B

nnsight transparent debugging

minTranscoders

Latent Space Clustering and Summarization

GuardianAI

Devising Effective Bechmarks

WELMA: Open-world environments for Language Model agents

CAMARA: A Comprehensive & Adaptive Multi-Agent framework for Red-Teaming and Adversarial Defense

ÆLIGN: Aligned Agent-based Workflows via Collaboration & Safety Protocols

General Pervasiveness

Demonstrating LLM Code Injection Via Compromised Agent Tool

Misinformational AI-Generated Academic Papers

GrandSlam usecases not technology

AI Agents for Personalized Interaction and Behavioral Analysis

RedFluence

BBC News Impersonator

Unsolved AI Safety Concepts Explorer

AI Research Paper Processor

Sleeper Agents Detector

VerifyStream

Web App for Interacting with Refusal-Ablated Language Model Agents

Alignment Research Critiquer

Alignment Research Critiquer

Data Massager

AI Alignment Toolkit Research Assistant

Reflections on using LLMs to read a paper

Academic Weapon

The House Always Wins: A Framework for Evaluating Strategic Deception in LLMs

Werewolf Benchmark

Can Language Models Sandbag Manipulation?

Deceptive behavior does not seem to be reducible to a single vector

Deceptive behavior does not seem to be reducible to a single vector

Detecting Lies of (C)omission

Detecting Deception with AI Tics 😉

Eliciting maximally distressing questions for deceptive LLMs

Evaluating Steering Methods for Deceptive Behavior Control in LLMs

An Exploration of Current Theory of Mind Evals

Sandbagging LLMs using Activation Steering

Towards a Benchmark for Self-Correction on Model-Attributed Misinformation

Boosting Language Model Honesty with Truthful Suffixes

Detection of potentially deceptive attitudes using expression style analysis

From Sycophancy (not) to Sandbagging

Gradient-Based Deceptive Trigger Discovery

Evaluating and inducing steganography in LLMs

Developing a deception dataset

Looking forward to posterity: what past information is transferred to the future?

Looking forward to posterity: what past information is transferred to the future?

Belief State Representations in Transformer Models on Nonergodic Data

Steering Model’s Belief States

Exploring Hierarchical Structure Representation in Transformer Models through Computational Mechanics

Detecting Anthropomorphic Tendencies in Language Models via Conversational Probing

LLM Benchmarking with Single-Agent Stochastic Dynamic Simulations

Benchmark for emergent capabilities in high-risk scenarios

WashBench – A Benchmark for Assessing Softening of Harmful Content in LLM-generated Text Summaries

Neurons and Attention Heads that Look for Sentence Structure in GPT2

Natural language descriptions for natural language directions

Observing and Validating Induction heads in SOLU-8l-old

Optimising image patches to change RL-agent behaviour

Regularly Oversimplifying Neural Networks

Sparsity Lens

Top-Down Interpretability Through Eigenspectra

Trying to make GPT2 dream

Visualizing the effect prompt design has on text-davinci-002 mode collapse and social biases

War is 15% conflic, 15% DragonMagazine