In a couple of Paul Graham’s essays about neural network computing he suggests that Semantic Stopsigns are a necessary part of the design for general-purpose, parallel-computing intelligences to keep them from getting stuck in infinite loops attempting to solve infeasibly large problems.
The key is learning to recognize it as an “overflow error” flag and not a “this problem is solved” flag. Internally they feel almost the same.
In a couple of Paul Graham’s essays about neural network computing he suggests that Semantic Stopsigns are a necessary part of the design for general-purpose, parallel-computing intelligences to keep them from getting stuck in infinite loops attempting to solve infeasibly large problems.
The key is learning to recognize it as an “overflow error” flag and not a “this problem is solved” flag. Internally they feel almost the same.