One question we need to answer before investing too much more effort into Sasquatch itself is whether Sasquatch is the right place for our ideas. An alternative would be to merge what you've done so far with LOTlib. They seem to have a well-developed library of examples, and the library itself also seems to be quite capable (i.e. can generate new kinds of abstractions from basic lambda calculus and even propose recursive functions). Several of the projects which interest us, however, remain to be tackled:
- I don't see SMT solvers or genetic programming as options for inference
- I don't see anything that explicitly learns anything approximating a new type, as opposed to simply learning a new function
- I don't see anything that seems to solve the problem of programming at multiple levels (i.e. working down from a high-level specification and up from a low-level language).
It might make sense to set up a meeting with Steve to see what they've been doing with LOTlib and where they're hoping to go with it. What do you think?
One question we need to answer before investing too much more effort into Sasquatch itself is whether Sasquatch is the right place for our ideas. An alternative would be to merge what you've done so far with LOTlib. They seem to have a well-developed library of examples, and the library itself also seems to be quite capable (i.e. can generate new kinds of abstractions from basic lambda calculus and even propose recursive functions). Several of the projects which interest us, however, remain to be tackled:
It might make sense to set up a meeting with Steve to see what they've been doing with LOTlib and where they're hoping to go with it. What do you think?