Are you going to try to evaluate it using Lossless Compression as per the
Hutter Prize and the Large Text Compression Benchmark?
If your AI actually predicts the answer - very accurately ("I think it's 80% a,
20% b") - to many diverse questions, then it will do extremely well at
compressing text, and etc. To make a general purpose solver you are finding
patterns, and finding patterns allows compression, we are making a multi
purpose hammer - not a one trick pony.
Testing your AI as you build it is crucial, it tells you if what you
implemented is working and if the parameters are approximately best. I run
compression hundreds of times, it absolutely guides you. It's as critical as
generated data. OpenAI exhibits both on their website. It's one thing seeming
funny to run but it is no joke it really works the best at compression and
really does tell you if you are making better AI.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T29d9a364b5df2085-M0dafc0954196ea2bcbfb9710
Delivery options: https://agi.topicbox.com/groups/agi/subscription
