Nick Cammarata, a researcher at Goodfire and former OpenAI employee, claims in an essay published on October 15, 2026 that the majority of neurons in large AI models can now be understood. According to his work using advanced agentic methods, nearly 90% of neurons in models like Qwen3-32B serve a single primary function, compared to only 1.7 to 2% in 2023 with the tools then available. Only 2 to 4% of neurons remain unexplained. These findings challenge the notions of representational superposition and polysemanticity that have shaped the field, although the scientific community’s reaction remains mixed.
Source: Read the original article
Disclaimer: this content is for information purposes only and is not financial advice. Cryptocurrencies are highly volatile: you may lose all of your capital. Always do your own research. Legal notice
