It's not a million to beat benchmark like your video title and it's only if you get over 85% in 12 hours of your notebook running during the contest. And it's distributed among the top entries, not all to one winner.
The ability to learn a new skill is not intelligence. Intelligence is many things but mainly applications of different types of reasoning then memorizing the results of that reasoning so you develop knowledge.
The point of the creator was that only truly intelligent systems can synthesize the rules necessary "on the fly" to solve it which is IMO not possible with transformer based models.
Brute forcing this proves nothing. Like the creator of the benchmark said only if you train your AI on general knowledge and rules and it still can then solve it it is a hint to intelligence. He also mentioned that it may be possible to train an transformer based model on this test but it wouldn't properly count as "solved" if it does which is true. If an AI can't solve this in a zero shot manner its not intelligent in my opinion (sure intelligence is very vague term but if you think an LLM is intelligent then a toaster or a calculater has its form of intellgience too according to that logic).
Great explanation, you just got another subscriber.
Great analysis and content
Great content Harshit !
It's not a million to beat benchmark like your video title and it's only if you get over 85% in 12 hours of your notebook running during the contest. And it's distributed among the top entries, not all to one winner.
So essentially they are asking for a full pledge JARVIS under 1 million USD and that too will be distributed partially.😂
The ability to learn a new skill is not intelligence. Intelligence is many things but mainly applications of different types of reasoning then memorizing the results of that reasoning so you develop knowledge.
The point of the creator was that only truly intelligent systems can synthesize the rules necessary "on the fly" to solve it which is IMO not possible with transformer based models.
no, fuck memorization, I want that thing generate solution on the fly without any training on previous examples
Brute forcing this proves nothing. Like the creator of the benchmark said only if you train your AI on general knowledge and rules and it still can then solve it it is a hint to intelligence. He also mentioned that it may be possible to train an transformer based model on this test but it wouldn't properly count as "solved" if it does which is true. If an AI can't solve this in a zero shot manner its not intelligent in my opinion (sure intelligence is very vague term but if you think an LLM is intelligent then a toaster or a calculater has its form of intellgience too according to that logic).