On Thursday, September 24, 2026, at 4:14 PM, Matt Mahoney wrote: > The Hutter prize has demonstrated training a 6M transformer with 4 bit > weights on one human level training set (1 GB) using 10^18 operations > including offline GPUs. I don't think this does human level inference yet > because information theory suggests needing 50M to 100M parameters to > represent the training data, which I estimate would need 10^19 operations > consuming 10^10 J or a few thousand kilowatt hours of electricity. A human > brain running at 20 watts for 75 years (2.5 x 10^9 seconds) uses 5 times more > energy. So I think the power problem can be solved with good engineering and > a few years of research.
There may be something occurring in the brain where encoding, storing, querying and transmitting information is different from the same old same old and LLM’s with their engineers haven’t figured out that efficiency yet. How do rare individuals who have had a brain injury wake up being able to speak different languages, play a musical instrument, or all of a sudden have hypercalculia and can recite Pi or compute primes? Prime fields are interesting: https://drive.google.com/file/d/1Jj1p9HmbasiC4747WxYsSu9a5ZWI-iZq/view ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T25fb5a030fcfa28e-Md2414b3245437377542890c2 Delivery options: https://agi.topicbox.com/groups/agi/subscription
