Grok 2, the latest AI model from xAI, has just been released, and I’m genuinely excited about it. This new iteration, along with its smaller counterpart Grok 2 mini, brings some interesting features to the table.
First off, Grok 2 was tested under the name ‘sus-column-r’ on LMSYS, and the results are impressive. The full-size Grok 2 performs exceptionally well, while Grok 2 mini, though not as powerful, still holds its own, especially in vision tasks.
One aspect I particularly appreciate is Grok 2’s integration with FLUX.1. This combination offers some great image generation. Even with Donald Trump in the image! It’s also worth noting that xAI has mentioned they’re working on their own image model, which could be better than FLUX, or a total flop. Who knows?
What sets Grok 2 apart for me is its refreshing approach to interaction. Unlike ChatGPT or Claude, which can sometimes come across as condescending, Grok 2 doesn’t shy away from political questions or controversial topics. Its system prompt, which was quickly extracted upon release, is amusingly straightforward with a phrase to the effect of: “Don’t be woke.” That’s all the political training it’s given. It’s such an Elon thing to do, and I find it oddly refreshing.
I’m eagerly waiting to see how Grok 2 will be priced in the API. Given its capabilities, it could be a strong contender in the AI assistant market.
It’s important to note that while Grok 2 isn’t open source, its predecessor, Grok 1, is. However, Grok 1 sucks and no one likes it. It is significantly less performant than models a fifth of it’s size.
Overall, I’m impressed with Grok 2. Its straightforward approach, combined with its powerful capabilities, makes it an LLM worth watching. Can’t wait for Grok 3!
