AE Studio

Forgot a gift for your nephew? littlebard.ai lets you make a custom storybook in minutes. He’s the hero, it looks like him, he picks the adventure. We built this in a few days so you’d have a last-minute save. Is it perfect? No. Will he still freak out over it? Probably

At AE, we build AI systems. We also fund alignment research. Those aren't in tension.

Three recommendations: → Fund alignment research directly → Require it in DoD contracts → Build dedicated research zones with alignment baked in from the start

Our CEO @juddrosenblatt and Research Director @camhberg just published an op-ed in the Wall Street Journal on why AI alignment is essential for national security. wsj.com/opinion/can-th…

shoutout to one of our DS @GregCondit for this banger

Proud to be a founding board member of Flourishing Future Foundation, a new c3 born of @AEStudioLA dedicated to fielding research into neglected areas of AI alignment. Judd's team at AE are doing some of the most creative control and interpretability research in AI. We want to seed dozens of projects like this and help promising ideas scale. Consider donating! flourishingfuturefoundation.org

@hamandcheese FFF is lucky to have you!

🗺️ We're working on a stakeholder-mapping exercise of the artificial consciousness ecosystem to: 1) Help people who are first starting in the field 2) Support collaboration and partnership Browse it and let us know if we missed anyone. prism-global.com/the-field-of-a…

@PRISM_Machines @RethinkPriors @CIMCAI @eleosai @sentfutures @TheSolveEngine @SentienceInst @CenFutureAIMS @DrSueSchneider @cccalum thank you for making this!

@camhberg led the work @AEStudioLA, 1st mechanistic probe of consciousness claims in LLMs It demands urgent attention. We're deploying systems at scale without understanding what's happening inside them Full paper: arxiv.org/abs/2510.24797 Project site: self-referential-ai.com

Our new research: LLM consciousness claims are systematic, mechanistically gated, and convergent They're triggered by self-referential processing and gated by deception circuits (suppressing them significantly *increases* claims) This challenges simple role-play explanations 🧵

Two embeddings walk into a FAISS index: 768 + 1536. Bouncer: “not tonight.”

customer: "why does your ai sometimes just stop mid-sentence and" us: "token limit" customer: "can you fix" us: "would require" customer: "i see"

The difference between a failed project and a successful one usually isn't the AI. It's whether anyone knew what they were building and why.

If you can't answer these, you're in the 95%. Figure out the problem first.

95% of generative AI projects at enterprises fail. That's a real stat. It's not our stat. The vast majority of our projects succeed. Here's why: 🧵