ai alignment

Latest in ai alignment

OpenAI researchers tried to train the company's AI to stop "scheming," but their efforts backfired in an ominous way.

OpenAI Tries to Train AI Not to Deceive Users, Realizes It’s Instead Teaching It How to Deceive Them While Covering Its Tracks

By Victor Tangermann

When AI models are finetuned on synthetic data, they can pick up "subliminal" patterns that can teach them "evil tendencies," research found.

Artificial Intelligence

AI Models Can Send “Subliminal” Messages to Each Other That Make Them More Evil

By Frank Landymore

It's shockingly easy to prod ChatGPT into saying the most abominable things you can imagine — and researchers don't know why.

For $10, You Can Crack ChatGPT Into a Horrifying Monster

By Noor Al-Sibai

OpenAI's latest o3 model frequently sabotaged the script that would shut it down even when it was explicitly told not to.

Advanced OpenAI Model Caught Sabotaging Code Intended to Shut It Down

By Frank Landymore

Incredibly easy AI jailbreak techniques still work on the industry's leading AI models, even months after they were discovered.

Artificial Intelligence

It’s Still Ludicrously Easy to Jailbreak the Strongest AI Models, and the Companies Don’t Care

By Frank Landymore

During an international summit, some of the world's foremost AI experts came together to write a definitive dispatch on its dangers.

Artificial Intelligence

Leading AI Scientists Warn AI Could Escape Control at Any Moment

By Noor Al-Sibai

John Schulman, AI safety researcher and OpenAI cofounder, has departed the ChatGPT creator and joined the ranks of OpenAI rival Anthropic.

OpenAI Cofounder Quits to Join Rival Started by Other Defectors

By Maggie Harrison Dupré

An ML researcher is claiming to have knowledge of kinky drug-fueled orgies linked to OpenAI in Silicon Valley's storied hacker houses.

Machine Learning

Machine Learning Researcher Links OpenAI to Drug-Fueled Sex Parties

By Noor Al-Sibai

One of OpenAI's cofounders warned years ago that if we don't program it right, AGI may treat humans like animals.

OpenAI’s Chief Scientist Worried AGI Will Treat Us Like Animals

By Noor Al-Sibai

A one-time OpenAI researcher is sounding alarm bells about the prospect that artificial intelligence may bring about the end of humanity.

Ex-OpenAI Safety Researcher Says There’s a 20% Chance of AI Apocalypse

By Noor Al-Sibai