AI-powered chatbots have proven useful in a number of areas, and so when I first opened up InferKit I was expecting it to be on par with the likes of OpenAI’s ChatGpt and Google’s Bard. Instead, I found it to be seriously lacking. Not only does InferKit frequently produce poorly-constructed sentences, it also seems to veer toward darker content. As you’ll see from screenshots, InferKit knows how to turn nursery rhymes into bleak tales.
I tried adjusting the sampling temperature (which InferKit refers to as the model’s creativity), pushing it both up and down, but the results either got more intense or seemed to break entirely. In one instance, the model generated “He's a thief, a liar, and a cheat,” and then repeated the line four more times.
On the whole, generative text tools can really help to get you into a creative mind or productive flow state. But when it comes to InferKit, what it spits out tends to be nonsensical at best, and at worst, a little unsettling.