New excellent update on ai news from excellent ai news aggregator ai search.
Edit: most impressive news is that the Nvidia avo harness together with claude managed to get 100% on arc agi 3 public benchmark, something that most models alone previously failed miserably at.
Edit 2: The fact adding a harness to claude allowed 100% on arc agi 3, when previously scores were under 30%, suggests that maybe llm technology with some additions might be able to reach agi, without having to resort to radically different architecture.
Intelligence might be simpler than some expected in its fundamental nature as Hawkins prior comments suggested.
"Intelligence and understanding started as a memory system that fed predictions into the sensory system. These predictions are the essence of understanding. To know something means that you can make predictions about it."- Jeff Hawkins
"[We need to see the brain] not as a fast processor, but as a memory system that stores and plays back experiences to help us predict, intelligently, what will happen next."- Jeff Hawkins
edit 3 A video by Michel Laclé titled
A Perfect Score: How NVIDIA's AVO Aced the Hardest Reasoning Benchmark
This short video focuses on the AVO harness and how together with claude it managed to ace the public arc agi 3 benchmark. Now it remains to be seen how it'll fare on the private arc agi 3 bench, but it will probably do well. The AVO harness was not designed for arc agi 3 yet managed such impressive results of 100%.
No comments:
Post a Comment