After feeding 270,000 words into a computer program that studies language patterns, The New Yorker got a result that’s about as awful as the real thing. The neural network studied each of these 270,000 words, as well as how they fit together contextually and came up with, well, this:
The preceding speech was written entirely by this AI language model. Unlike human speech, The New Yorker could dial the program up from one to five, each with its pros and cons. Level one, for example, is fairly accurate speech, albeit with lots of repetition. At level five, we see the program work at full capacity, but what comes out is, believe it or not, actually more nonsensical than Trump’s actual speech.
In my estimation, the real Donald Trump falls somewhere between levels one and two. There’s lots of repetition, a few made up words, and a lot of random nonsense. On his better (worse?) days, he could even approach a level three, but only at times.
Interesting, but as the video proves: we’re not quite there yet.
Get the TNW newsletter
Get the most important tech news in your inbox each week.