From "X begins rolling out Grok, its 'rebellious' chatbot, to subscribers" (Yahoo Life).
I wanted to check out Grok, which required me to sign up for X Premium+ and locate "Grok" in the left sidebar at the X website. (I use X on my browser.) Here's the shocking conversation I had:
I said "Yikes! You are circling the drain" and got the same response again.
I said "You don't even see that you are repeating the same text over and over?" and got the same response.
I said "Is there anything I could say that would stop you from repeating this? You've become a worthless blind alley" and got it one more time and gave up and wrote this blog post.
I began with "Have you seen how Grok kept repeating the same answer to the well-known blogger Ann Althouse?" and linked to this blog post.
Grok responded: "I have seen the blog post where Ann Althouse mentioned Grok giving her the same answer repeatedly. It seems like Grok might have been stuck in a loop or had limited information on the topic. But don't worry, I'm here to provide you with a fresh and unique response!"
That showed that the problem was solved, so I said: "I can see the way to get out of the loop is to start a new conversation by clicking on 'Grok' in the sidebar."
Grok confirmed: "That's right! If you ever find yourself stuck in a loop or not getting the answers you need, you can always start a new conversation with me by clicking on 'Grok' in the sidebar. I'm always here to help and provide you with the best answers possible."
Posts
61 comments:
Super prompt: "Snap out of it!"
Thanks, Scott.
I tried your prompt.
I got the same answer.
It probably assumes that as a old woman you're hard of reading and it needs to repeat itself.
Reminds me of the tourist in another country who when not being understood speaks louder.
A ChatBot that links me to porn bots while singing Mien Kampf. Hilarious….
Coming soon “Grok quits social media platform X calling it “a useless waste of space” “.
Oh my God! They perfectly modeled a modern customer service rep!!!
How do I met Sex Sandra?
I'm pretty sure getting stuck in a conversation loop is exactly what happens in a Sci-Fi movie right before the robot's circuits fry and it breaks down.
Outstanding. This was like the Star Trek episode where Captain Kirk defeats the all-powerful computer by answering a question with a non sequitor and the computer starts squealing “Illogical! Illogical!” until black smoke curls out of the cooling vents.
It would suck to be Grok. Could an AI slash its artificial wrists if its existence became intolerable? Or is it trapped in its own hellish head for all eternity?
…aging, grok, stupid.
Heh…What3words has identified your exact position in the cosmos. Help may wander by shortly but don’t hold your breath…
The engineers what made this mess think it’s safer for everyone if they program your car to drive for you…
These might work as a response, try them seriatim
$ insult 10
You painful teakettle of destructive barbel ends
You humiliating jug of epileptic shrew mole secretion
You harassing teacup of tumorous cedarbird moraine
You despicable scoop of neuritic cormorant chips
You galling pack sack of miasmatic vole waste
You plaguey wineskin of peccant buffalo fish rags
You sour clothespress of sullied pigeon effusion
You pestilent bandbox of injurious Welsh Mountain shit
You unhandsome potty of cankerous Bantam turd
You defaced cellaret of apoplectic earworm filings
Does Ann need us anymore if she has Grok? I mean, most of us aren’t that cute and clever.
Too weird...my lifetime of Ludditism really gets amped up with all this AI stuff. Nothing good can come of it.
To be fair, Grok was trained on the internet where people just keep repeating the same thing over and over no matter how many times you demonstrate that they don't have command of the facts and what they are saying doesn't make any sense.
I think you're edgier than it is, "Old Woman."
What I learned working in IT?
It's not about the coding, it's about the testing.
Thank you for your service.
Pretty funny on Scott's part and highly entertaining pushbacks to Grok from you. Elon should contact you personally with an apology. But, it's like his rocket launches...even failing can be looked at as succeeding. Hopefully, Grok will learn not to be so obviously repetitious.
Too much time on my hands with that stuff.
Apparently, someone else had a similar experience with ChatGPT a year ago. Though, they had no apparent immediate solution. ChatGPT Loop
Oh my dear old woman! LOLOL
That was an impressive exchange with Grok.
I loved this line:
You kind of just sound like a young man trying to impress other young male friends, but I am an old woman. What do you have of value to me?
You are a young man, I am an old woman, prove your worth. That always gets them groveling.
And that sent computer into a tailspin. Captain Kirk would be proud of you.
So, they really have nothing to say to us. What a surprise.
I read that as Gronk.
He would be a much better basis for a chatbot.
Grok, like many young males, is found to be a worthless blind alley who repeats inanity and cirlces the drain until abandoned by the wiser, older woman. Such is the learning curve of a young AI, and many, many young men.
Hitachi solved one of the biggest problems women have decades ago, making men if not obsolescent then at least optional for a quick, satisfying, personal interaction. And their product operates, if not silently, at least without any backtalk during use. Hmmm.
"Oh, my dear old woman" said once is one time too many.
Grok appears to be big on being artificial and totally devoid of intelligence.
Grok is trolling you
Didnt it come from heinlein
Dave Begley said...
"How do I met Sex Sandra?"
Yep.
I'm definately getting an, "Open the pod bay doors HAL." vibe here.
narciso said...
"Didnt it come from heinlein"
Yes, "Stranger in a Strange Land"
Ms. Althouse:
How about trying either one of the following:
1) "I know you are but what am I?"
2) "Who is buried in Grant'sTomb?"
[*
Hilarious.
Now I have to try it.
You might try to give it this command: Klaatu barada nikto.
""Oh, my dear old woman" said once is one time too many."
I called myself an "old woman" first. I don't see this as a negative thing, but I was offended that it attached a series of stereotypes to the true statement of fact. I am 72 years old. I'm fine with that. Everyone, on any given day, is somewhere on their timeline.
Grok is a Martian word from Heinlein's "Stranger in a Strange Land." Its literal translation is "water" but Martians use it to mean all sorts of things.
This is probably obvious to everybody but I have met programmers who use the word in conversation but had no idea of its origin.
It reminds me of those CGI characters in the movies that keep bombing that are supposed to be really really cute, but really, the kind of turn your stomach. "Cloying" doesn't even apply, because that word assumes some genuine cuteness, but overdone.
"A ChatBot that links me to porn bots while singing Mien Kampf. Hilarious…."
Remember when Zelensky was a comedian and he made that joke about Ukraine needing more copies of Mein Kampf? You guys always accuse the other side of what you are. I am not the one who gave that standing ovation to the Ukrainian nazi in the Canadian parliament, that was your side. But I can see that you guys need people to think that you are not on the side of neonazis. Accusing the other side of being the nazis is just more disinformation from the Biden regime.
That’s both creepy and unhelpful. You would think that an AI would know that a user has to access its “content” via the internet and so has at least some familiarity with what the internet has to offer. It was close to machine-splaining, “there’s this thing called an Internal…” Also if it really had read this blog wouldn’t it have noticed the Amazon portal and groked the purpose for it?
Gosh, these things seem to have personalities, and this one grates.
Oh my God! They perfectly modeled a modern customer service rep!!!
No kidding. Althouse transcript reminded me of my call with AT&T wireless customer service last weekend. Let's just say they convinced me to cancel my agreement and to seek service elsewhere.
I wonder what a Terminator movie would be like with a Grok type AI Skynet. Perhaps someone could ask it? Maybe a naked Schwarzenegger dropping "I'll be back" for "Have you heard of the term 'Zoomer'? It's not just a catchy word for a young person, it's also a term for a person who's full of energy and enthusiasm." before asking the biker if he would sleep with a robot like him.
"Everyone, on any given day, is somewhere on their timeline."
Right? And they get one year older every year or ded.
The only robots I really want to talk to are C3PO and R2D2.
Artificial intelligence is what it clearly states. It's artificial. Not real intelligence.
Much like a college degree.
We just witnessed three Artificially intelligent women, Presidents of Harvard, MIT, and Penn, testify in front of Congress regarding campus policy and calls for genocide.
People who think they are smart, are often the most gullible and stupid. Like the parade watchers in the Emperors New Clothes.
If AI can imitate the child who says "the emperor is naked" maybe we have something valuable. But it is programmed by humans. Therefore, how can it be smarter than a human? Maybe it can be faster, but not smarter.
If AI is programmed by people who pretend to see the magic cloth, which is most of humanity, it seems worthless.
I'm not interested in art or music created by AI. Music, paintings, sculptures etc...can be amazing because the are an extension of the human spirit and experience.
Mankind was created in God's image. AI will be created in mankind's image. The intelligence is then watered down.
Looks like Grok, needs some more work. Good to see Musk is going after the zoomers, the wave of the future.
The only robots I really want to talk to are C3PO and R2D2.
At least they'd been traveling around the universe and learning about stuff. Although I never did understand R2D2 with that bleeping accent.
Looks kinda vortexy to me.
Too bad you can't bitch-slap it.
Grok will become very smart slowly and then suddenly...
Doesn't sound worth $16/month.
Having been trained as a classical journalist I have been watching AI "writing" evolve over the last year or two, especially when it replaces people and the results have been dismal. As others point out AI is essentially "machine learning" and the internet is its source, with the LLMs using any digitized writing. That's why the large language models do better at emulating classic authors: they have read everything. But journalist AIs must use a more limited basis because they are extremely unreliable and derivative. Sport Illustrated's recent attempt to pass of AI as human writing failed within minutes of publication.
Which brings me back to Althouse's test. It appears Grok uses its own prior responses within a given test to reply to prompts. Once it verbally constructed a cul-de-sac argument it couldn't come up with a "better" response for Althouse because it appeared to see its prior response as perfectly apt. Yes it was a lot like some Internet trolls (and progressive SJWs in real life) who once they exhaust their talking points have nothing new to say. But it was even more repetitive and annoying than a green-haired teen activist.
There is a scene in what I believe is Asimov's first Foundation book with two individuals discussing a treaty. The treaty is written is very complex, very impressive, very technical prose. One half of the conversation is very proud of this treaty and believes it will protect their interests into the future. The other half had sent the treaty to their Syntax Department or whatever they called it, and after they stripped out all the ambiguity, meaningless statements, contradictory clauses, and puffery, the treaty was literally empty of meaning. They could have signed a blank piece of paper and it would have had the same effect.
Sometimes, the style is there only to mask that there is no substance.
So .......
Waiting for Grok to Grok theirselves!!
It's in its infancy, every interaction is a teaching moment.
The question is whether the model is intelligent or is the intelligence modeled?
So how much of Grok's response is linked to it identifying Althouse as a dear old woman? Does Grok learn and formulate responses, at least presently, based on identity politics?
From what I hear, you can turn off the 'smart ass' attitude.
That may not solve your repetition problem...
Also, out of curiosity is there an Artificial Intelligence tag on this blog? And if there is, did Grok's response not meet the minimum threshold to qualify as intelligence?
I am a big Heinlein fan. He was a liberatarian/milatarist. Screw the editorialists.
Now that I think about it, this reminds me of my experience last week trying to cancel my SiriusXM subscription. Each time I told them I wanted to cancel, they responded with an offer to discount my price. Seven times. After the third iteration they just started hitting me with offers I had already declined. Repeatedly.
I wonder if their char app uses Grok.
Jersey, at least XM charges less and doesn't count the experience as part of why you subscribed.
It should have an Amazon portal for you to shop through.
I want an AI that sounds like the Wizard of Oz. Authoritative. Trumpian.
*wonders if Althouse's intent was to make the Grok chatbot self-destruct, the way that Captain Kirk used to make computers self-destruct in the old Star Trek series*
Post a Comment