The dumbest AI taught the smartest AI. Here’s how that went…
This video is about weak-to-strong generalization: whether a weaker AI can successfully teach a stronger AI. This is important for superintelligence alignment, because humans may eventually need to supervise AIs that are smarter than they are. If weak supervisors can help align stronger AIs, then humans (or future AIs helping humans) might be able to align superintelligence. In this video, we explore OpenAI’s experiments on this question in depth. Read the paper here: https://arxiv.org/abs/2312.09390 Robert Miles AI Safety: @RobertMilesAI AI Safety courses by BlueDot Impact: https://bluedot.org/ ▀▀▀▀▀▀▀▀▀PATREON, MEMBERSHIP, MERCH▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀ 🟠 Patreon: / rationalanimations 🔵 Channel membership: / @rationalanimations 🟢 Merch: https://rational-animations-shop.four... 🟤 Ko-fi, for one-time and recurring donations: https://ko-fi.com/rationalanimations ▀▀▀▀▀▀▀▀▀SOCIAL & DISCORD▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀ Rational Animations Discord: / discord Reddit: / rationalanimations X/Twitter: / rationalanimat1 Instagram: / rationalanimations TikTok: / rational.animations BlueSky: https://bsky.app/profile/rationalanim... ▀▀▀▀▀▀▀▀▀PATRONS & MEMBERS▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀ Thanks to all our patrons and channel members from the Simple Adder tier and above! A Alcher Black Alexander230 Amir Saboury Apuis Retsam blasted0glass Bleys BlueNotesBlues Chad M Jones Chris Painter Christian Loomis Craig Falls Danealor Daniel Chica Danilo Stefani - Alessandra Erba David Piepgrass Dawson Ducky Ed Edward Yu Ellis Jones Felix Akkermans Forodriac Origamius Fraser Cain Gabriel Ledung Glenn Tarigan Honyopenyoko Ingvi Gautsson Ivan Bachcin Jackson Emanuel James Babcock Jana JanJan Jasper L Jeroen De Dauw joe39504589 John John Everett-Slape Juan Benet Klemen Slavic Kristin Lindquist loopuleasa Luke Freeman Matias Badino Michael Andregg Michael Hewitt Michael Reed Nathan Fish Nathan Metzger Neal Strobl NMS noggieB Odet Abadia Patryk Wielopolski rictic Robert Paul Schwin Scott Alexander Sequoia SQRT42Pi steven michaels Stuart Alldritt Terberlo.dog Tomas Campos Tor Barstad ttw Vladimir Silyaev Zachary Taylor Arjun Arul John 7ic7ac Thomas Grip Teo Val Jotunus Torstein Haldorsen BestProGaming Rinthean Arthur Petron dangered wolf Laissez Scholar Boris Bend Ken Mc AWyattLife ▀▀▀▀▀▀▀CREDITS▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀ The extremely muscular team that made this video happen: https://docs.google.com/document/d/1E...

AI could be a tool for global control (plus other major AI risks)

The story of Omega-L and Omega-W

What Do Neural Networks Really Learn? Exploring the Brain of an AI Model

That Alien Message

Humanity was born way ahead of its time. The reason is grabby aliens.

Every Math Paradox Explained

AI Sleeper Agents: How Anthropic Trains and Catches Them

Math's Most Deceptive Sequence

I Gave ChatGPT a Body

We let AI buy a robot and a car, it does exactly what experts warned.

This May Be Humanity’s Hardest Challenge

How to Take Over the Universe (in Three Easy Steps)

Why Young People Are Rejecting AI

But how do AI images and videos actually work? | Guest video by Welch Labs

Will we grab the universe? Grabby aliens predictions.

The True Story of How GPT-2 Became Maximally Lewd

ASMR Mysterious Growth ❓ CLOSE Medical Exam 👩⚕️Professional Doctor Facial Examination

Oligarchy is worse than you think

What If you Could See the 4th Dimension?

