4 ms·
https://evanthebouncy.github.io/program-synthesis-minimal/ https://evanthebouncy.github.io/program-synthesis-minimal/ Here's my take on it, you can view it as
by evanthebouncy 2y ago
https://evanthebouncy.github.io/program-synthesis-minimal/ https://evanthebouncy.github.io/program-synthesis-minimal/
Here's my take on it, you can view it as a modern extension to Armando's work (he's my PhD advisor)
- algo_trader 2y agoFor practical systems, to improve synthesis/codegen, do you think working on the compile/eval toolchain is a good direction ? For example, if we can cache/re-use the compiler run on previous codegen candidates, to speed up compilation of the next candidate sniplet
- evanthebouncy 2y agoeval is crucial, as that's the only way to agree with the computer what it is doing matches what you want it to do. I wouldn't worry about speeds. We should expect compute and inference to be faster in the future, where we can sample easily 10k programs in a second, check all of them against test cases. I'd worry about communicating precisely to computers when test cases are awkward to write.