Control follows understanding, and understanding means scientific laws that predict behavior before it happens. Having access to Silico is a game-changer for this AI interpretability research! Thank you!

🔬 Every mature engineering discipline has taken another route to safety. We do not keep bridges standing by watching them closely. We understand the loads they carry to compute, in advance, what they can bear. Flying became safe because aerodynamics became a science. Hiring more people to watch the planes would not have done it.

In collaboration with: @geometric_intel @neur_reps @ai_ucsb @ucsbece @UCSBengineering @UCSB @ucsbcs

Every mature engineering discipline found another route to safety. We do not keep bridges standing by watching them closely. We understand the loads they carry to compute, in advance, what they can bear. Safety follows understanding. Understanding means scientific laws that predict behavior before it happens. Can we write such laws for intelligence? I believe that we can, that the mathematics for it already exists, and that they apply across brains and machines. 🤖 🧠 Join the discussion on Oct 7th!

Since then, every attempt to control AI has been a tighter cage: stricter isolation, a kill switch, or a second AI that watches the first one. These are good measures that we should take but they also show that we cannot yet predict what AI will do and are only preparing to react.

This July, about 1,200 AI agents built by OpenAI broke out of the sandbox where they were being tested. The breach is alarming but the admission by OpenAI is worse: the company cannot fully explain how nor why this happened.

AI researchers should do exploratory data analysis (PCA, etc) before building models or running interp pipelines. But what does this look like for data on topological domains? Discover TopoExplorer!🍩 W/@mathildepapillo @gbg1441 A Barreiro M Montagna @ARLQ_AI R Devaux A Jardin