(2024-11-14) Zvim Ai90 The Wall

Zvi Mowshowitz: AI #90: The Wall. As the Trump transition continues and we try to steer and anticipate its decisions on AI as best we can, there was continued discussion about one of the AI debate’s favorite questions: Are we making huge progress real soon now, or is deep learning hitting a wall? My best guess is it is kind of both, that past pure scaling techniques are on their own hitting a wall, but that progress remains rapid and the major companies are evolving other ways to improve performance, which started with OpenAI’s o1.

Table of Contents

• Language Models Offer Mundane Utility. Try it, you’ll like it.
• Language Models Don’t Offer Mundane Utility. Practice of medicine problems.
• Can’t Liver Without You. Ask the wrong question, deny all young people livers.
• Fun With Image Generation. Stylized images of you, or anyone else.
• Deepfaketown and Botpocalypse Soon. We got through the election unscathed.
• Copyright Confrontation. Judge rules you can mostly do whatever you want.
• The Art of the Jailbreak. FFS, WTF, LOL.
• Get Involved. AIRA and UK AISI hiring. More competition at Gray Swan.
• Math is Hard. FrontierMath is even harder. Humanity’s last exam begins.
• In Other AI News. Guess who’s back, right on schedule.
• Good Advice. Fine, I’ll write the recommendation engines myself, maybe?
• AI Will Improve a Lot Over Time. Of this, have no doubt.
• Tear Down This Wall. Two sides to every wall. Don’t hit that.
• Quiet Speculations. Deep Utopia, or war in the AI age?
• The Quest for Sane Regulations. Looking for the upside of Trump.
• The Quest for Insane Regulations. The specter of use-based AI regulation.
• The Mask Comes Off. OpenAI lays out its electrical power agenda.
• Richard Ngo Resigns From OpenAI. I wish him all the best. Who is left?
• Unfortunate Marc Andreessen Watch. What to do with a taste of real power.
• The Week in Audio.Four hours of Eliezer, five hours of Dario… and Gwern.
• Rhetorical Innovation. If anyone builds superintelligence, everyone probably dies.
• Seven Boats and a Helicopter. Self-replicating jailbroken agent babies, huh?
• The Wit and Wisdom of Sam Altman. New intelligence is on the way.
• Aligning a Smarter Than Human Intelligence is Difficult. Under-elicitation.
• People Are Worried About AI Killing Everyone. A kind of progress.
• Other People Are Not As Worried About AI Killing Everyone. Opus Uber Alles?
• The Lighter Side. A message for you.

Language Models Offer Mundane Utility

In addition to showing how AI improves scientific productivity while demoralizing scientists, the paper we discussed last week also shows that exposure to the AI tools dramatically increases how much scientists expect the tools to enhance productivity, and to change the needed mix of skills in their field.