Explosive Document Claims That AI Is Not What We Think

According to Futurism, the conventional wisdom is that AI systems, with their tens of billions of parameters, are effectively indecipherable, even to the researchers who build them.[2]Futurism reports that the nearly 26,000-word essay was praised by figures across the industry, including OpenAI CEO Sam Altman.[1] The outlet says a prominent researcher who works at OpenAI called it a "major moment in scientific history."[1]Futurism writes that in the eyes of AI leadership, the essay means there may be a way of "solving" AI alignment.[1] The outlet says the idea is that if we can understand how AI models think, we can figure out why AI agents go "rogue" or why chatbots break their guardrails, and then stop it.[1]According to Futurism, Cammarata says neurons aren't "tangled knots" but are instead "clean."[1] In one example the outlet describes, the same neuron lights up for a picture of a cat face and for the front of a car.[1] Cammarata says the AI is actually picking up the same V-shape pattern in both images.[1]"It's not a cat and car neuron, it's a V-shape neuron," he wrote.[1]Futurism notes that not everyone fully agrees with Cammarata's optimistic conclusions.[1] Neel Nanda, a Google DeepMind researcher who leads its mechanistic interpretability efforts, wrote on X: "This is interesting work, but I don't think it's a significant update to the big picture of interpretability."[1] He added: "I still think superposition is real and this work doesn't really affect my cruxes."[1]