Grok AI Hallucinates Less Than Any Other AI. Believe It or Not!
January 6, 2026
Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.
I had to put down my Telegram Notes project to capture the factoids in “Elon Musk’s Grok Records Lowest Hallucination Rate in AI Reliability Study.” Of course, I believe everything I read on the Internet. I am a veritable treasure trove of the factoids about Elon Musk’s technologies; for example, full self driving, the trade in value of a used Tesla Model S, and the launch date for a trip to Mars. I am so ready.
The AI system tells the programmers, “Yes, you can use AI to predict which of the egame players are most likely to have access to a parent’s credit card. This card can be used to purchase tokens or crypto to enter higher-stake gambling games. Does that help?” Thanks, Qwen. Too bad you weren’t in the selection of AI systems tested by the wizards at Relum.
The write up presents information from gold-standard sources too; for instance, the casino games aggregation company Relum. And what did this estimable company’s research reveal? Consider these factoids:
- Grok’s hallucination “rate” is just eight percent. Out of 100 prompts / queries, Grok goes off the rails only eight times. This is outstanding. Everyone wants such a low rate of hallucination. Exceptions may apply for some nitpickers.
- The worst LLMs in the hallucination rate category are ChatGPT and Google Gemini. These outfits make up information more than one third of the time. That’s not too bad if you are planning on selling adds. The idea is “prompt relaxation.” The more relaxed, the wider the net for allegedly relevant ads. More ads yield more revenue. Maybe there is more to making up answers than meets the idea. I am okay with ChatGPT and Google competing for the most hallucinogenic crown. Have at it, folks.
- Deepseek, the Chinese freebie, hallucinates only 14 percent of the time. Way to go, Chinese political strategists. (Qwen’s hallucination was not reported in the article. By the way, that’s one of the models Pavel Durov’s Telegram will allegedly rely upon to translate Messenger content and perform other magical functions for the TONcoin pivot. Note the word “magical.” Two public companies listed on NASDAQ in six months. As I said, “Magic.”
Here’s a quote from the gambling company’s chief product officer. Obviously this individual is an expert in the field of machine learning, neural networks, matrix transforms, and the other bits and bobs of building smart software. Here’s the statement:
About 65% of US companies now use AI chatbots in their daily work, and nearly 45% of employees admit they’ve shared sensitive company information with these tools. These numbers show well how important chatbots have become in everyday work.
Absolutely. When one runs Windows, the user is “using” smart software. When one uses Google, AI is there. These market winners are moving forward, greased on wheels of fabricated output. Yeah, great.
Several observations:
- Grok seems unaware of messages posted on X.com. I wonder why.
- A bad actor has to sign up to access the Grok API and the X.com API in order to pull off some slick AI-based cyber activities. I wonder why.
- Grok’s appeal to online gaming companies is interesting. I wonder way.
I have no answers. Relum does. These data do not reassure me about Mr. Musk’s business tactics for building Grok’s market footprint.
Stephen E Arnold, January 6, 2025
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
