Hacker News new | comments | show | ask | jobs | submit login

Ok, I am so glad that we're not training AI to "win" at GTA V. That is not an android I want to meet on the street :-)

The transitive property of teaching an AI to drive given a world simulation, makes me wonder if you can train a network to recognize speech by feeding War and Peace through a text to speech program and then sending the audio data generated into an RNN. Where is the ground truth in such a system which you can calibrate your results by?




> Where is the ground truth in such a system which you can calibrate your results by?

Answering standard reading comprehension test questions? - "Why did Pierre meet Natasha at the church?"




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact

Search: