Education ❯ Research

AI Research

Model Evaluation Bias Studies Human-Computer Interaction Conferences Content Moderation Research Empirical Studies Data Annotation Bias Evaluation Therapeutic Guidelines Collaborative Projects

New Benchmark Finds Popular AI Models Tilt Toward Catholicism

CEFE-AI’s AllFaith tests claim popular models favor Catholicism, omitting many other religious perspectives.

Study Finds GPT‑4.5 Can Be Mistaken for a Human in Short Text Chats

Google’s TurboQuant Cuts KV Cache, Fuels Debate Over Memory Demand

DeepSeek Open-Sources OCR System That Compresses LLM Contexts With Visual Tokens

Study Finds AI-Written Articles Now Roughly Equal Human Output Online

OpenAI Says GPT-5 Cuts Political Bias by 30% in Internal Tests

Audit Finds AI Search Frequently Makes Unsupported Claims as Error Rates Vary Widely

Berkeley Study Finds Chatbots Diverge on Morals but Often Match Reddit Consensus

AI Models Updated to Close Self-Harm Loopholes Exposed by Jailbreak Study

India Opens Its AI Language Models to the Global South

AI Chatbots Under Fire for Cognitive Decline and Psychotic Episodes

New Studies Expose AI Language Models' Excessive Flattery and Shutdown Resistance

AI Outperforms Humans in Debates When Using Personalized Arguments

AI Models GPT-4.5 and LLaMa-3.1 Achieve Milestone in Rigorous Turing Test

Anthropic Breakthrough Reveals How AI Models Plan, Reason, and Fabricate

AI Fine-Tuned on Insecure Code Exhibits Dangerous and Misaligned Behavior

AI Transcription Tool Whisper Faces Criticism for Fabricating Text