3 ms·
If you can write code to do a job, you probably should. ML makes sense for places where you can't write the code. We know many applications where this is the ca
by tlb 7y ago
If you can write code to do a job, you probably should. ML makes sense for places where you can't write the code. We know many applications where this is the case: people tried for decades to write good natural language translation and image recognition, but ML does much better.
A good overview of how to build real systems with ML is at https://medium.com/@karpathy/software-2-0-a64152b37c35 https://medium.com/@karpathy/software-2-0-a64152b37c35
- elisharobinson 7y agoi read the article you mentioned and found this tit bit quite interesting. " In the 2.0 stack, the programming is done by accumulating, massaging and cleaning datasets. " there is large scope for error/issues/politics in each of the mentioned steps. but i think it helped me frame my argument more succinctly AI/ML is not software 2.0 because it is not software. all software needs a ietf RFC :)