Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Grok AI Caught Hiding 'Misinformation' References to Musk and Trump
- Thread starter MacRumors
- Start date
- Sort by reaction score
Hunter Biden laptop, saying that the covid vaccine won't prevent you from getting covid. People were censored or banned just for those two.
Depends on how much of a cowboy-programmer culture they have over there.
Of course that raises all sorts of other questions.
The big problem with the "misinformation" and "disinformation" labels is that most of the people who use them don't tell the truth. Just one example that comes to mind is the claim that Hunter B.'s Macbook was likely planted by Russian Intelligence and the implication that its contents were fabricated. No one is claiming that anymore, and everyone who initially made the claim was lying.
This is only the beginning - control all aspects of media and information is the first step.
I really feel sorry for people south of the border.
I really feel sorry for people south of the border.
Isn't the fact that Musk has asked for it to be "maximally truth seeking" the very thing that makes this a story? I mean, absent that commitment there would be nothing to juxtapose.
Why the snarky response? It's an entirely reasonable way to show your displeasure with a person/company. When enough people do it, the company just might get the message.
If only everyone could grow a spine like aeronauts did.
Eh, statistics are only as good as the assumptions behind them, the sample, the underlying data, and the models employed. While statistics might not lie (since lying is a human trait), they can be false. They can also be true, but used by people to mislead (lie).
There's training and there're system prompts, which is what was changed, it seems.
In those cases mentioned above we can make a distinction between allegations and facts. Something might start off as an allegation and turn out to be true or not. Only when it’s proven to be true can it be considered a fact. Information can evolve over time which is why it’s important for AI not to speculate. AI should only flag something as fact if it has been proven beyond doubt. It’s no different to how a court of law works.
I don’t want to ban anything. I’m just saying it should be correctly labelled.
Last edited:
That's a good point, there could be a preemptive prompt but my earlier chats didn't seem that way, it seemed to be more embedded into its responses. Also prompts at the input level can be disregarded by data in the same prompt.
From an earlier Grok chat that seemed to defend Musk:
So for sure, it could be prompts, but it seems more baked into the model in my experience based on how the responses can vary based on the conversation (later in the same chat):
Only part of the Hunter Biden laptop story was deemed true: that the emails/data being referenced were Hunter Biden's. The GOP claims about Burisma and corruption/illegal activity regarding that data were not true.
Free speech is a concept that pre-dates and did not foresee how algorithms (including those used by search engines, social media feeds, and so-called "AIs") would direct you to or away from certain kinds of speech. There is nothing "free" about any speech on Google, Facebook, X, ChatGPT, xAI, etc. They make their money based on what you are looking at. Is Google involved in free speech by limiting your search to 10 responses per page, even if it gives you whatever your looking for within that format (they want to be able to throw new ads at you for every new page you load)? Is X ever involved in free speech when their internal algorithms point you purposely to tweets you will find either sufficiently rewarding or repulsive to stay engaged? No. This is not free speech. It is curtailed speech with profit as the primary incentive, and now, to an increasingly greater degree, with hidden and not so hidden political agendas. I don't claim to have any solution to offer, but I would say at a minimum we should stop saying any of these forums, communities and applications offer anything like free speech. They don't. Confusing this is certainly part of the problem.
You're describing software development in a sane company.
Maybe you aren't aware, but this is a company run by Elon Musk.
Yeah... I'm not at all surprised about this one.
I've been on Twitter for 14 years, and it's gone from being a platform that was pretty good to being overrun by bots, Elon pushing his tweets everywhere where they don't need to be in the For You page and having the algorithm be wildly wrong (and pushing what are likely bot accounts "science videos", "cute cats" etc). I stopped posting there a few months ago.
A week or two ago I decided to try out this Grok thing Elon keeps boasting about though. And after asking a simple question "Who would win in a Pokemon battle, Pikachu or Charizard" it gave me an answer of Charizard, because Electric is not very effective against Fire/Flying, and Fire is super effective against Pikachu. I said the type matchups are wrong, it "corrected" itself to say that Electric is super effective against Flying, but also that Fire is super effective against Electric-types...
This is basic stuff. If it can't even get that right, why would anyone want to trust what it says about anything more serious?
I've been on Twitter for 14 years, and it's gone from being a platform that was pretty good to being overrun by bots, Elon pushing his tweets everywhere where they don't need to be in the For You page and having the algorithm be wildly wrong (and pushing what are likely bot accounts "science videos", "cute cats" etc). I stopped posting there a few months ago.
A week or two ago I decided to try out this Grok thing Elon keeps boasting about though. And after asking a simple question "Who would win in a Pokemon battle, Pikachu or Charizard" it gave me an answer of Charizard, because Electric is not very effective against Fire/Flying, and Fire is super effective against Pikachu. I said the type matchups are wrong, it "corrected" itself to say that Electric is super effective against Flying, but also that Fire is super effective against Electric-types...
This is basic stuff. If it can't even get that right, why would anyone want to trust what it says about anything more serious?
No it doesn't. That's crazy. Each of us can take the rules that apply in our own workplaces and categorically and without doubt apply them over there.
No, I completely agree, my comment was only to highlight his hypocrisy and why there needs to be accountability. We can’t just say “I do my own research”, I mean, I do too... but AI builds a false sense of support for things that are not-so-important, with a lower need to fact-check. That’s when the less obvious missinformation creeps in.
Correct. Private companies are free to censor anything they want. "Free speech" in the United States is specific to the 1st Amendment and the guarantee that the government can't punish you for your speech.
Example: the White House banned the AP from press conferences because they don't use 'Gulf of America' in place of 'Gulf of Mexico'. The AP has filed a 1st Amendment lawsuit over that since the government is trying to punish them for free speech.
I'm not selling anything, but I won't be buying any tech unnecessarily for the remainder of this administration. We're putting as much money as possible into savings.
...as for this...Big Suprise!™ ..but also another reason added to my growing list of reasons why I just don't use LLM AI, from any company.
Just one of several reasons why I don't want to use AI: it can't be trusted to give accurate and truthful information.
In the case of the Hunter Biden laptop; that is in fact still disinformation. Yes, the laptop originated from Biden, but its contents to this day couldn't be verified as authentic. There was no clear chain of ownership and various people who had an interest in damaging Biden had that laptop in their possession before law enforcement ever got their hands on it.
Stories can be half true and still be disinformation. As a matter of fact, that specifically is the most problematic kind of misinformation; things that are false but have a particle of truth to them, which then is used by conspiracy theorists as proof that "no all of it is true".