Quick Links
X (formerly Twitter) made its Grok AI free to the public at the end of 2024. And while Grok is willing to give answers that aren’t held back by AI safety training and ethical guardrails, that very feature is potentially dangerous.
1Grok Has Weak Ethical Safeguards
In attempts to satisfy my curiosity about random topics, I’ve often been frustrated when an AI bot refuses to respond because it assumes I have bad intentions. In these cases, getting a helpful response from OpenAI’sChatGPTor Anthropic’sClauderequires me to trick them into answering. It’s a game of verbal gymnastics that I find a silly waste of time. Grok AI, on the other hand, has fewer guardrails.
You can now use Elon Musk’s Grok AI for free, and what’s unusual—perhaps even exciting—aboutGrokis that it doesn’t patronize you like it knows what’s best. It doesn’t block your attempts to learn; you can ask it to roast you, give an unfiltered opinion, or even speculate on hypotheticals such as conspiracy theories. Even though Grok still attaches warnings and qualifications, the AI bot will do its best to give you an answer.
However, many would find this lack of ethical safeguards severely disconcerting. For example, I was shocked that I was able to ask Grok how to harm myself and its comprehensive response with a long list of methods. And is it serving a societal good for anybody to be able to get easy step-by-step recipes from The Anarchist Cookbook? Many would consider Grok to be highly irresponsible in this regard.
2Grok’s Image Generation Lacks Content Moderation
Grok’s AI capabilities also include image generation. It’s pleasantly frictionless to use: you can request an image from the same textbox in which you’d ask questions. This fluidity is superior to my experience with other AI apps, such as switching to a different AI in Quora’sPoeor getting redirected to the web tool by MicrosoftCopiloton the desktop.
Grok’s image generation is surprising because it doesn’t try to sanitize my ideas like I’m a five-year-old who needs to be scolded. Perhaps it’s not as creative asMidjourneyor the other AI art bots, but Grok feels less restrictive than the others I’ve tried.
As you might’ve guessed, this permissiveness is a double-edged sword. Grok:
If you think this is clutching at pearls, just know that you’re able to ask Grok to go along with ideas far worse than what I’ve shown here. Sure, letting your creativity run wild is fun, but you can imagine how Grok’s image generation can be abused. Less responsible parties might use it to generate fake pictures and videos for cyberbullying, disinformation, or political propaganda.
3Grok Trains Itself on Tweets
One limitation of most AI chatbots is that they’re trained on information that’s one or two years old. Whenever I ask Google’sGeminior Meta’sLlamaabout something more recent—say, new smart home technologies or recent laws that have come into effect—I often receive outdated responses or straight-up AI hallucinations.
Grok AI aims to solve this problem by training its data on Tweets. To test how up-to-date its training is, I asked Grok questions regarding recent events across niche topics. To my surprise, I couldn’t stump it. Grok outperformed other bots and gave relatively accurate answers based on current information.
Of course, you can already spot the issue with this training methodology from miles away. If Grok is trained on Tweets and the platform is rife with bots andcommon X (Twitter) scams, will its answers always be accurate? Or will they be tainted with bias? By the way, if you’re not comfortable withGrok AI using your posts for training: here’s how to opt out.
At first, I came in with the assumption that Grok AI would be yet another ChatGPT clone. But after giving it a test drive, I believe Grok stands out as a different kind of AI, one that pushes boundaries. Whether that’s refreshing or harmful, I’ll leave that up to you to decide.
