flâneur

13 Arguments About a Transition to Neuralese AIs — LessWrong

lesswrong.com · 5,544 words · saved by 1 readers

Over the past year, I have talked to several people about whether they expect frontier AI companies to transition away from the current paradigm of transformer LLMs toward models that reason in neuralese within the next few years. This post summarizes 13 common arguments I’ve heard, six in favor and seven against a transition to neuralese AIs. The following table provides a summary: Following Hidden Reasoning in LLMs: A Taxonomy, I’ll define neuralese as reasoning in long opaque chains of recurrence. For a precise operationalization of this, I like Fabien Roger’s definition and his corresponding prediction market. He defines the current, CoT-based reasoning paradigm as the conjunction of the following two properties: The ability to perform 100k heavy serial operations without text bottlenecks already seems like a strong deviation from the current paradigm, where reasoning models perform up to 500 serial operations in a forward pass.[1] Thus, in this post, I’ll constrain the number of a

x 13 Arguments About a Transition to Neuralese AIs — LessWrong Aether AI Frontpage 52 13 Arguments About a Transition to Neuralese AIs by Rauno Arike 7th Nov 2025 11 min read 14 52 Over the past year, I have talked to several people about whether they expect frontier AI companies to transition away from the current paradigm of transformer LLMs toward models that reason in neuralese within the next few years. This post summarizes 13 common arguments I’ve heard, six in favor and seven against a transition to neuralese AIs. The following table provides a summary: Arguments for a transition to neu

related reading