Rules and text generation

0:00 In this video, I'll compare all of them, show you what they can and can't do, and how to use them to their fullest. This will be an unusual comparison. I won't just go over facts or compare the responses to the same prompt. Instead, I'll compare my experience with them, the little things I noticed, and I'll assign points, you know, rating. In each category, I'll have things to add points for for consistency in images, quality, research, etc.

0:47 For example, for the same feature, one tool can get nine points because it's almost perfect, another one five points, and another tool zero. Then I'll sum them up and announce the winner. Thank you Hrefs for sponsoring this video. The GBT isn't officially out yet. That's true, but all the leaks point to the same idea behind it. Taking all those models we have right now and combining them into one. No separate model for reasoning or for text generation or for working with large context inputs.

1:09 One model to roll them all. In this idea, Grock is the closest one right now. You only have one model and all the toggles for research or reasoning only turn on the features themselves without changing the model. Gemini does the same thing but needs to switch a model for research. And Chad GBT right now, you know how it is. So, right off the bat, I'll give Grock two points, bonus points, and Gemini one point for outpacing Chad GBT.

1:32 The number one skill of all LLMs is writing. They all do it really well, but only if you start looking at small things, you start noticing the rift between them and different approaches they chose. Take style for example. All three can slip into any voice you want. Shakespeare, newspaper reporter, or a hyper 5-year-old. The first reply is usually perfect. The trouble starts when you drop in a long chunk of text or shout change style halfway. That is when they wobble.

1:54 Chad GBT is best here.