3 ms·
Language models don't just repeat, they have randomness in their outputs linking synonyms together. That's why their output can be novel and isn't just plagiari
by naasking 4y ago
Language models don't just repeat, they have randomness in their outputs linking synonyms together. That's why their output can be novel and isn't just plagiarism. How this might translate to code isn't entirely clear.
- tiborsaas 4y agoTransformers were first intended to be used for translation. To them code is just another language. Code is much more rigid than a human language so I think it's not that surprising that it can produce custom code.