> Gizmodo reached out to Grok-developer xAI for comment, but they have only responded with the usual automated reply, “Legacy Media Lies.”
European here, so perhaps not my place to have an opinion on domestic U.S. legal policies, and I don't want to make this political (although I guess it kind of is…) BUT:
Why are no media outlets on the offense when companies use these kinds of statements? Shouldn't Gizmodo, or its owner Keleops Media, treat this as slander and take it to court? If Grok's behavior can be objectively verified, why is it so easy for a company to get off the hook so easily just by saying "lies" and move on?
Are photos & videos that new even recognizable by Grok? Can't try ChatGPT because it spews an error ("Error in message stream") and then I've reached my limit. Gemini on gemini.google.com (Gemini 3 Pro) does the same as Grok and thinks the photo[1] is from the October 7th attacks:
>Based on the visual evidence and public reports, the man in the image is Yarden Bibas.
>His identity and background have been widely confirmed by news organizations and his family following the events of October 7, 2023.
On Google AI Studio (Gemini 3 Pro with Grounding enabled) it correctly recognizes him as Ahmed Al-Ahmed. In the Chain-of-Thought it first thinks it's from the Oct 7th attacks:
>I'm now investigating potential misidentification. The name "Ahmed Al-Ahmed" is a lead, but the visual suggests recent events. [...] However, the visual clues still point towards a recent conflict. The "thank you" narrative implies gratitude, sparking the idea of a Bedouin man rescuing Israelis during the Hamas attack. I'm actively using Google to explore these new avenues.
>[...]
>I've got a match! Ahmed al-Ahmed, the Bondi Beach hero, is confirmed. The snippets consistently mention December 14-15, 2025, and today's the 15th! He fits the description: Syrian refugee shop owner, tackled a gunman at a Hanukkah event, and was shot. The visual confirms a man in a white shirt being helped, with a message of thanks. This is definitely the right event!
So no "misinformation" or "glitching", just LLMs being LLMs.
It is so irresponsible to call this a "glitch". This isn't something going wrong at any point, this is just technology that is not sufficient for what it is being asked to do, and only because we know the real truth and are able to fact check does that register.
Claiming it is a glitch gives the opportunity for AI companies to hide behind the excuse of "mistakes in the code", instead of recognizing the fundamental flaw of the technology in question.
At the same time, this article attempts to politicize a wider issue by relating the failings of AI to current events. In fact, this hallucination and failure is near constant, but it is no coincedence. It is the product of both technology being used before readiness, and the (hilarious) attempt by Elon to use AI as a propaganda machine to spread and legitimize his beliefs.
18 comments
[ 3.4 ms ] story [ 38.4 ms ] threadEuropean here, so perhaps not my place to have an opinion on domestic U.S. legal policies, and I don't want to make this political (although I guess it kind of is…) BUT:
Why are no media outlets on the offense when companies use these kinds of statements? Shouldn't Gizmodo, or its owner Keleops Media, treat this as slander and take it to court? If Grok's behavior can be objectively verified, why is it so easy for a company to get off the hook so easily just by saying "lies" and move on?
>Based on the visual evidence and public reports, the man in the image is Yarden Bibas.
>His identity and background have been widely confirmed by news organizations and his family following the events of October 7, 2023.
On Google AI Studio (Gemini 3 Pro with Grounding enabled) it correctly recognizes him as Ahmed Al-Ahmed. In the Chain-of-Thought it first thinks it's from the Oct 7th attacks:
>I'm now investigating potential misidentification. The name "Ahmed Al-Ahmed" is a lead, but the visual suggests recent events. [...] However, the visual clues still point towards a recent conflict. The "thank you" narrative implies gratitude, sparking the idea of a Bedouin man rescuing Israelis during the Hamas attack. I'm actively using Google to explore these new avenues.
>[...]
>I've got a match! Ahmed al-Ahmed, the Bondi Beach hero, is confirmed. The snippets consistently mention December 14-15, 2025, and today's the 15th! He fits the description: Syrian refugee shop owner, tackled a gunman at a Hanukkah event, and was shot. The visual confirms a man in a white shirt being helped, with a message of thanks. This is definitely the right event!
So no "misinformation" or "glitching", just LLMs being LLMs.
[1] https://x.com/NoaMagid/status/2000196326521204984
I see this sooo soooo much but folks will just straight up ask “@grok is this true?” and its response it taken as gospel.
Though I have to say, grok code-fast-1 is one of the best coding models I’ve ever used.
Claiming it is a glitch gives the opportunity for AI companies to hide behind the excuse of "mistakes in the code", instead of recognizing the fundamental flaw of the technology in question.
At the same time, this article attempts to politicize a wider issue by relating the failings of AI to current events. In fact, this hallucination and failure is near constant, but it is no coincedence. It is the product of both technology being used before readiness, and the (hilarious) attempt by Elon to use AI as a propaganda machine to spread and legitimize his beliefs.
What kind of a news story is that? It’s an LLM. It serves thousands of requests per minute. Of course some answers will be silly.