3 ms·

I was thinking about this lex friedman podcast with Marcus Hutter. Also, Joshua Bach defined intelligence as the ability to accurately model reality.. is lossl

• akomtu 2y ago
  It's the ability to find a simple model that predicts a complex reality with a high accuracy and low latency. So we need to consider the four axes, and AI will be a region in this space. Edit. In fact, there is a simple test for intelligence: can you read a function in C and tell how a change in input changes the output. For complex algorithms, you have to build an internal model because how else are you going to run qsort on million items? That's also how you'd tell if a student is faking it or really understands. A harder test would be to do the opposite: from a few input/output examples come up with an algorithm.
• meowface 2y ago
  Incidentally, François Chollet, creator of ARC-AGI, argues in this 2020 Lex Fridman podcast that intelligence is not compression: https://youtu.be/-V-vOXLyKGw https://youtu.be/-V-vOXLyKGw
  • visarga 2y ago
    Compression works on finite data, and AI models need to keep open for new data. So they should be less greedy.
• fastball 2y ago
  For a quicker tie in than watching a whole podcast, the result of Hutter's stance takes the form of the Hutter prize[1], which in some ways has many of the same goals as ARC-AGI, but sees compression itself as the goalpost towards intelligence. [1] http://prize.hutter1.net/ http://prize.hutter1.net/