Use Grok (Update your Gen AI Awareness #1)

Grok 3 Beta from Twitter/X is one of the best models available for free

Last week, I pointed a bunch of things you need to do to become a Gen AI expert. And also, that you don’t have a choice, you have to become a Gen AI expert.

One problem with most people is that they tried out ChatGPT (or continue to use it in with a free account) and they haven’t really caught on to the fact that the world has changed, and free ChatGPT is now pretty bad (compared to the other options that are available).

So I’m starting a series of short posts, called “Update your Gen AI Awareness”, with byte-sized information about what has changed in the world of Gen AI while you were not paying attention.

One big change is that Grok 3 Beta, the new AI model available from xAI (Elon Musk’s AI company), is one of the best models available to free users today.

Some quick points:

• ChatGPT free account users mostly get GPT 4o, which is increasingly bad compared to the other good models. A paid ChatGPT account is absolutely worth it for ChatGPT 4.5, Deep Research, o3-mini-high, and o1. But 4o is only to be used for quick-and-dirty questions, especially if they involve web searches.

• Claude 3.7 Sonnet is also available to free users and is one of the best models around—if your answer doesn’t require searching the web for latest (or obscure) information. Also, you only get a limited number of 3.7 Sonnet uses after which it downgrades you to the Haiku model which is not as good.

• Grok is free, and free users are allowed a lot of usage of Grok 3 Beta, which is comparable to some of the best models available today. And, as far as I can tell, the number of free uses of Grok 3 Beta is higher if you have a Twitter account (even if it is not a premium account).

• Grok is the best at accessing latest information, breaking news, and so on. As expected, it does a good job of looking up information in Twitter and Reddit and summarizing it for you. For example, I was considering buying a Nissan car and asked it about Nissan’s after-sales service record in my city, and it did a better job of summarizing the results than the other models and search engines.

• Grok is also better at generating images than ChatGPT/DallE. The latter is just broken in many ways.

If you don’t understand a tweet (a rather common complaint a lot of my friends have about my tweets) just click on the Grok button at the top left of a tweet and you’ll get a lovely explanation from grok. Like below:

Grok explains my tweets so that I can become even lazier in my tweeting.

• Grok is integrated with Twitter: I frequently find myself clicking the Grok button to understand certain tweets or historical references or scientific terms that are new to me. Many more people need to start doing this more often (and also using the Meta.AI integrated in WhatsApp; but more on that in a later post)

• Important warning: Grok has a much higher chance of hallucinating (giving wrong information with confidence) and of biased responses compared to other comparable models. Of course, you have to always be careful of errors and biases in all Gen AI usage, but this is far more true in the case of Grok.

Think of it this way: your team of a dozen experts has just hired (for free!) a new team member who is very smart but is also overconfident—you usually get excellent answers but sometimes it is convincing bullshit. Remembering to use Grok at the right time (and double-checking its work) is the new challenge you need to master this week. Next week will be a different challenge.