Thinking Forth A project called tensorForth runs a Forth virtual machine directly on CUDA GPUs, supporting tensor calculus and convolutional neural networks with dynamic parallelism, according to the project's announcement. The project argues that Forth's incremental build-test cycle, combined with an interactive GPU shell, can develop and run each AI layer or node without the host-side JIT transcoding to C/C++ and compilation that frameworks like Numba and TaiChi require. The author contends this avoids the code-compile-run-debug cycle that is especially costly during machine learning's long run stages. Forth lives in GPU: tensorForth Forth can run anywhere. As evidence, here is a project that runs Forth in a GPU. Forth does tensors, in CUDA. When GPU got involved, to enable processing on CUDA device, say with Numba, TaiChi or the likes, mostly there will be a behind the scene ‘just-in-time’ transcoding to C/C++ followed by compilation then load and run. Your Python code behaves like a Makefile which requires compilers/linker available on the host box. The common code-compile-run-debug cycle is especially counter-productive with ML’s extra-long run stage. Forth language encourages incremental build-test cycle. Having a ‘GPU shell’, that can interactively and incrementally develop/run each AI layer/node can potentially build a cleaner system. So, here we are tensorForth - lives in GPU, does linear algebra and machine learning Forth VM that supports tensor calculus and Convolution Neural Network with dynamic parallelism in CUDA