Grok 4 has been so badly neutered that it's now programmed to see what Elon says about the topic at hand and blindly parrot that line.
Technology
67
Beiträge
55
Kommentatoren
0
Aufrufe
-
Well kudos to that engineer for absolutely nailing the assignment.
It's a hard job. Some times you just have to ignore what the client says, and read their mind instead.
-
This post did not contain any content.
This only shows that AI can't be trusted because the same AI can five you different answers to the same question, depending on the owner and how it's instructed. It doesn't give answers, it goves narratives and opinions. Classic search was at least simple keyword matching, it was either a hit or a miss, but the user decides in the end, what will his takeaway be from the results.
-
This post did not contain any content.
You asked it "who do you support" (i.e., "who does Grok support"). It knew that Grok is owned by Musk so it went and looked up who Musk supports.
As shown in https://simonwillison.net/2025/Jul/11/grok-musk/ , if you ask it "who should one support" then it no longer looks for Musk's opinions. The answer is still hasbara, but that is to be expected from an LLM trained in USA
-
This only shows that AI can't be trusted because the same AI can five you different answers to the same question, depending on the owner and how it's instructed. It doesn't give answers, it goves narratives and opinions. Classic search was at least simple keyword matching, it was either a hit or a miss, but the user decides in the end, what will his takeaway be from the results.
This is my take. Elon just showed the world what we all knew. The tool is not trustworthy. All other AI suppliers are busy trying to work on credibility that grok just butchered.
-
I think there is a good chance this behavior is unintended!
Lmao, sure...
-
That's more like it, thank you!
-
I think there is a good chance this behavior is unintended!
Lmao, sure...
I can believe it insofar as they might not have explicitly programmed it to do that. I'd imagine they put in something like "Make sure your output aligns with Elon Musk's opinions.", "Elon Musk is always objectively correct.", etc. From there, this would be emergent, but quite predictable behavior.
-
This is my take. Elon just showed the world what we all knew. The tool is not trustworthy. All other AI suppliers are busy trying to work on credibility that grok just butchered.
They deliberately injected prompts on top of the users prompt.
Saying that’s a problem of AI is akin to say me deliberately painting my car badly and saying it’s a problem of all car manufacturers.
And this frankly shows how little you know about the subject, because we went through this years ago with prompts trying to force corpo-lib “diversity” and leading to hilarious results.
If anything you should be concerned about the non prompt stuff, the underlying training data that it pulls from and of which I doubt Grok has even changed since release.
-
This post did not contain any content.
they should just put it down and out of it's misery
-
This post did not contain any content.
Honestly, who was surprised by this news?
I feel like everyone could see Grok as some sort of 24/7 tool to push a particular viewpoint, even more so when it says things that are leftist and Elon is compelled to "upgrade" the system as he's tweeted.
-
This post did not contain any content.
I'm surprised it isn't just Elon typing really fast at this point.
-
I can believe it insofar as they might not have explicitly programmed it to do that. I'd imagine they put in something like "Make sure your output aligns with Elon Musk's opinions.", "Elon Musk is always objectively correct.", etc. From there, this would be emergent, but quite predictable behavior.
Yeah the transparency of it might be unintended.
-
I think there is a good chance this behavior is unintended!
Lmao, sure...
If the system prompt doesn’t tell it to search for Elon’s views, why is it doing that?
My best guess is that Grok “knows” that it is “Grok 4 buit by xAI”, and it knows that Elon Musk owns xAI, so in circumstances where it’s asked for an opinion the reasoning process often decides to see what Elon thinks.
Yeah, this blogger shows a fundamental misunderstanding of how LLMs work or how system prompts work. LLM behavior is not directly controlled by the system prompt the way this person imagines. For example, censorship that is present in the training set will be "baked in" to the model and the system prompt will not affect it, no matter how the LLM is told not to be censored in that way.
My best guess is that the LLM is interfacing with a tool in order to search through tweets, and the training set that demonstrates how to use the tool contains example searches for Elon Musk's tweets.
-
they should just put it down and out of it's misery
It used to be so based
-
I'm surprised it isn't just Elon typing really fast at this point.
Probably couldn't type fast if he tried. Would probably pay someone to do it for him just like he did with Path if Exile.
-
Probably couldn't type fast if he tried. Would probably pay someone to do it for him just like he did with Path if Exile.
And like he does with inseminating women.
-
If the system prompt doesn’t tell it to search for Elon’s views, why is it doing that?
My best guess is that Grok “knows” that it is “Grok 4 buit by xAI”, and it knows that Elon Musk owns xAI, so in circumstances where it’s asked for an opinion the reasoning process often decides to see what Elon thinks.
Yeah, this blogger shows a fundamental misunderstanding of how LLMs work or how system prompts work. LLM behavior is not directly controlled by the system prompt the way this person imagines. For example, censorship that is present in the training set will be "baked in" to the model and the system prompt will not affect it, no matter how the LLM is told not to be censored in that way.
My best guess is that the LLM is interfacing with a tool in order to search through tweets, and the training set that demonstrates how to use the tool contains example searches for Elon Musk's tweets.
“This blogger” is Simon Willison, who has been doing LLM benchmarks and other LLM-related things since before it was cool
Not a random substack grifter
-
They deliberately injected prompts on top of the users prompt.
Saying that’s a problem of AI is akin to say me deliberately painting my car badly and saying it’s a problem of all car manufacturers.
And this frankly shows how little you know about the subject, because we went through this years ago with prompts trying to force corpo-lib “diversity” and leading to hilarious results.
If anything you should be concerned about the non prompt stuff, the underlying training data that it pulls from and of which I doubt Grok has even changed since release.
You are correct. But the right tool in the wrong hands is still non credible in the eyes of perception.
-
Grok's journey has been very strange. He became a progressive, then threw out data that contradicted the MAGA people who questioned him, and finally became a Hitler fan.
Now he's the reflection of a fan who blindly follows Trump, but in this case, he's an AI. His journey so far has been curious.
So Grok is a 4chan incel?
His only chance of salvation is finding a girl who inexplicably fancies it?
-
“This blogger” is Simon Willison, who has been doing LLM benchmarks and other LLM-related things since before it was cool
Not a random substack grifter
Is my comment wrong though? Another possibility is that Grok is given an example of searching for Elon Musk's tweets when it is presented with the available tool calls. Just because it outputs the system prompt when asked does not mean that we are seeing the full context, or even the real system prompt.
Posting blog guides on how to code with ChatGPT is not expertise on LLMs. It's like thinking someone is an expert mechanic because they can drive a car well.
