By default I actually don't trust anyone's ML code. It's just very easy to make subtle mistakes which make things just slightly worse. But @ikostrikov is one of the very few exceptions I have. I'll blindly trust his implementations. It's no surprise he's part of this 👇 result!
By default I actually don't trust anyone's ML code. It's just very easy to make subtle mistakes which make things just slightly worse. But @ikostrikov is one of the very few exceptions I have. I'll blindly trust his implementations. It's no surprise he's part of this 👇 result!
@giffmana @ikostrikov Do you trust your own implementations though 🤔
@giffmana @ikostrikov making worse to trust.. human being hard to trust and now their developed ML too.. to err is human.. to err is code.. OMG
@giffmana @ikostrikov They are using jax of course. I can get similar results with TF graph these days but I would rather keep up with everyone else. Also makes it harder for people to trust my code I am sure. But I work on windows primarily and its not going to be so easy.
@giffmana @ikostrikov DNNs are very powerful, have lots of capabilities in learning non-linear mappings that are too complex. So, there could be mistakes within the code that may not elucidate themselves as everything seems to be in working order (e.g. accuracy, latency, etc.) Good to be skeptical !