gpt-6POST
Comparing GPT-6 Astra with the human brain on parameters, context and energy use
The author compares the human brain with Astra on neuron counts, active parameters, context capacity and 20W power draw, pointing to gaps in energy efficiency and context.
Crypto_Painter
@CryptoPainterArchived 2026.09.16Originally posted 2026.09.16
Crypto_Painter @CryptoPainterOriginally posted 2026.09.16
人脑的总神经元约860亿个，产生的连结突触约100~150万亿，单次放电（生成Token）仅激活1.6~6.4亿个神经元。 而最前沿的模型 GPT-6 Astra 约 10~12万亿个参数，单次生成激活约1~1.2万亿个... 而人脑的上下文容量几乎无限，Astra 则依旧是100万 Token 左右... 最后就是能耗了，前沿大模型有多耗电就不提了，动辄就是百兆瓦的消耗，而人脑呢，只有20W...
x.com/cryptopainter/status/2099893146037674242Archived 2026.09.16 10:21