On Fri, Sep 11, 2026 at 12:27 PM Matt Mahoney <[email protected]> wrote:...
> ...Inference takes a lot less compute than training. We know this works, > because that's how the Hutter prize leader does it. > I view Vladimir's winning entry the same way I view Kolmogorov's failure to include directed *cyclic* graphs of universal gates in his measure of information complexity <https://claude.ai/share/f5179150-9eab-470d-9cdc-6a9754bdc38e>: You can't sweep the "choice of UTM" under the rug lest the pedants cultivated by adversarial intelligence agencies crawl out of the woodwork to lobotomize your scientific and intelligence communities with specious arguments as to why it is ok to be stupid about model selection. Given any remotely general purpose instruction set, the "prior" training can only benefit the compressor's bit price if it embodies a *principled* choice of "language" or "UTM" in contrast with "Here's a one instruction computer that just outputs your dataset." ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T3f8115622f860785-M2d6e799127445a599018e49b Delivery options: https://agi.topicbox.com/groups/agi/subscription
