Yes of course, something practical to test, but a better pattern finder is much more accurate than something weaker but more practical. 6 weeks and/or 128GB RAM would be much more worth it than ex. a 1 day & 4GB algorithm once they both reach fully trained (approx.).
You don't need to test on enwik9 or even enwik8, you can do enwik7 and still see the winner. Now you can practically test the stronger AI. And yes trained full for application will be 6 weeks BUT now we know it's worth it. So we still have a problem here with Mr. Hutter, the true contest should use enwik7 no? No difference there. And should not restrict speed/memory, the fast AI will take longer because although it may eat 1,000 times more data than the slow strong AI it'll need trillions of trillions of trillions to be as accurate. What is his reasoning? ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T6761a13445e5864b-M82c9d170d9548a31222a64b6 Delivery options: https://agi.topicbox.com/groups/agi/subscription
