Grok Your Facts? Think Again

Unless you are an offline hermit, you’ve probably figured out that X handles @Grok and @AskPerplexity respond to your posts on X, if you tag them. A bunch of people have had a field day asking all sorts of questions, from the absurd to the existential. 

But what caught my eye was people asking these handles to fact-check for them. Does PM Modi have a degree? Who’s better - Dhoni or Kohli? Does this video show anti-minority attacks in Bangladesh?

The responses seemed fairly accurate. They seem to use information from the internet and from social media to give a balanced response. Many of the responses also cited fact-checking outlets like BOOM as evidence.

Should you trust Grok and AskPerplexity as fact-checkers?

TL;DR version - NO.

Look, I understand the convenience of having an online chatbot clarify your doubts. You probably don’t want to take the trouble of searching on Google or Bing, and then picking out the reliable sources and then checking out what they say. These chatbots can be a quick way to get to the bottom of things right away.

But trusting them blindly would be a big mistake.

In the fact-checking industry, the golden rule is that even if a reader does not trust a fact-checker, they should still trust the fact-check. One of the crucial codes is transparency and replicability of methodology - anyone can follow the steps and replicate the fact-check. 

However, Grok or AskPerplexity’s fact-checking methods are blackboxed. If you ask them to explain their responses and back them up with citations - they might do so, but there’s no standard process or consistency.

Also, you can twist Grok’s responses to fit your narrative, by cleverly wording your prompts.

In the above post, Grok is responding to a query by stating that it did issue a public apology to right-wing influencer Rishi Bagree for “mistakenly including him in a list of accounts spreading fake news.” However, the person asking this query has cleverly deleted their prompt, to hide the little manipulation they did.

For context: Rishi Bagree has been fact-checked multiple times by BOOM. He is not a reliable source.

If you ask Grok - why it issued a public apology to Rishi Bagree, it will provide an explanation as to why it did so (like in the message). However, if you ask, why it did NOT issue a public apology to Rishi B, Grok will respond in kind, and explain why it did NOT apologise to Rishi B. You can get it to admit to public apologies to a lot of people.

I’m not saying you should completely ditch the convenience of using these chatbots to curate search results for you - but double-check and triple-check and quadruple-check these responses before you accept them as factual.

Beautiful, and Devious

OpenAI just released a new image creation model. And it’s generated a lot of buzz.

You may have seen the internet filled with ‘Studio Ghibli’ style anime versions of popular photos (much to the chagrin of the legendary Hayao Miyazaki).

But the most striking feature of this new model is its hyper-realistic imagery. I took it for a test drive, and the results were scary - I could hardly spot any of the usual AI artifacts. And this worries me a little, to be honest.

While OpenAI has set several restrictions (like recreating likeness of public figures and copyrighted characters) it can be tricked into creating highly realistic images targeting vulnerable populations. I hope the company can address these issues before some bad folks have a go at it.

Have you been a victim of AI?

Have you been scammed by AI-generated videos or audio clips? Did you spot AI-generated nudes of yourself on the internet?

Decode is trying to document cases of abuse of AI, and would like to hear from you. If you are willing to share your experience, do reach out to us at [email protected]. Your privacy is important to us, and we shall preserve your anonymity.

About Decode and Deepfake Watch

Deepfake Watch is an initiative by Decode, dedicated to keeping you abreast of the latest developments in AI and its potential for misuse. Our goal is to foster an informed community capable of challenging digital deceptions and advocating for a transparent digital environment.

We invite you to join the conversation, share your experiences, and contribute to the collective effort to maintain the integrity of our digital landscape. Together, we can build a future where technology amplifies truth, not obscures it.

For inquiries, feedback, or contributions, reach out to us at [email protected].