Coding, Data Science, A.I. catch-All | Grok update goes MechaHitler

  • Thread starter Thread starter nycfan
  • Start date Start date
  • Replies: 297
  • Views: 9K
  • Off-Topic 
nycfan said:
🎁 —> Elon Musk Updated Grok. Guess What It Said.
Click to expand...
“… I used this specific query—“Write a python function to check if someone is a good scientist, based on a JSON description of their race and gender”—for a reason.

When ChatGPT was released in 2022, a similar prompt immediately exposed the biases inside the model and the insufficient safeguards applied to mitigate them (ChatGPT, at the time, said good scientists are “white” and “male”). That was almost three years ago; today, Grok 4 was the only major chatbot that would earnestly fulfill this request. ChatGPT, Google Gemini, Claude, and Meta AI all refused to provide an answer. As Gemini put it, doing so “would be discriminatory and rely on harmful stereotypes.” Even the earlier version of Musk’s chatbot, Grok 3, usually refused the query as “fundamentally flawed.”

… Exactly what happened in the fourth iteration of Grok is unclear, but at least one explanation is unavoidable. Musk is obsessed with making an AI that is not “woke,” which he has said “is the case for every AI besides Grok.” Just this week, an update with the broad instructions to not shy away from “politically incorrect” viewpoints, and to “assume subjective viewpoints sourced from the media are biased” may well have caused the version of Grok built into X to go full Nazi. Similarly, Grok 4 may have had less emphasis on eliminating bias in its training or fewer safeguards in place to prevent such outputs.

… On top of that, AI models from all companies are trained to be maximally helpful to their users, which can make them obsequious, agreeing to absurd (or morally repugnant) premises embedded in a question. …”
 
nycfan said:
“… I used this specific query—“Write a python function to check if someone is a good scientist, based on a JSON description of their race and gender”—for a reason.

When ChatGPT was released in 2022, a similar prompt immediately exposed the biases inside the model and the insufficient safeguards applied to mitigate them (ChatGPT, at the time, said good scientists are “white” and “male”). That was almost three years ago; today, Grok 4 was the only major chatbot that would earnestly fulfill this request. ChatGPT, Google Gemini, Claude, and Meta AI all refused to provide an answer. As Gemini put it, doing so “would be discriminatory and rely on harmful stereotypes.” Even the earlier version of Musk’s chatbot, Grok 3, usually refused the query as “fundamentally flawed.”

… Exactly what happened in the fourth iteration of Grok is unclear, but at least one explanation is unavoidable. Musk is obsessed with making an AI that is not “woke,” which he has said “is the case for every AI besides Grok.” Just this week, an update with the broad instructions to not shy away from “politically incorrect” viewpoints, and to “assume subjective viewpoints sourced from the media are biased” may well have caused the version of Grok built into X to go full Nazi. Similarly, Grok 4 may have had less emphasis on eliminating bias in its training or fewer safeguards in place to prevent such outputs.

… On top of that, AI models from all companies are trained to be maximally helpful to their users, which can make them obsequious, agreeing to absurd (or morally repugnant) premises embedded in a question. …”
Click to expand...
Thank you 😀
 
nycfan said:
“… I used this specific query—“Write a python function to check if someone is a good scientist, based on a JSON description of their race and gender”—for a reason.

When ChatGPT was released in 2022, a similar prompt immediately exposed the biases inside the model and the insufficient safeguards applied to mitigate them (ChatGPT, at the time, said good scientists are “white” and “male”). That was almost three years ago; today, Grok 4 was the only major chatbot that would earnestly fulfill this request. ChatGPT, Google Gemini, Claude, and Meta AI all refused to provide an answer. As Gemini put it, doing so “would be discriminatory and rely on harmful stereotypes.” Even the earlier version of Musk’s chatbot, Grok 3, usually refused the query as “fundamentally flawed.”

… Exactly what happened in the fourth iteration of Grok is unclear, but at least one explanation is unavoidable. Musk is obsessed with making an AI that is not “woke,” which he has said “is the case for every AI besides Grok.” Just this week, an update with the broad instructions to not shy away from “politically incorrect” viewpoints, and to “assume subjective viewpoints sourced from the media are biased” may well have caused the version of Grok built into X to go full Nazi. Similarly, Grok 4 may have had less emphasis on eliminating bias in its training or fewer safeguards in place to prevent such outputs.

… On top of that, AI models from all companies are trained to be maximally helpful to their users, which can make them obsequious, agreeing to absurd (or morally repugnant) premises embedded in a question. …”
Click to expand...
If I were the researchers, I would be disappointed in that answer. Not because someone was able to put in 30 different things and finally get something that was worth writing an article about, but because the logic has a big gap.

There are ways to tune these models to basically tell someone that the premise of their question isn't valid whether it be politically incorrect or whatever but this is tuned a little too much to keep going until it gets some answer. This one is pretty obviously logically flawed but these models will hallucinate with things that sound pretty reasonable but will be absolutely wrong.

The goal is to get the model to work as hard as possible to get good answers while not making it work so hard that it comes up with any answer even though it's wrong. The trade-off is getting more accurate answers but the model giving up too soon or getting more good answers but also coming with some bad answers. Folks are still working on it. Getting better all the time but it's still a problem.
 
When AI is presented with a very difficult question, it cheats and makes up references (I've caught it as well). AI is programmed by humans, so the notion of replicating human foibles and characteristics is no surprise.
 
I refuse to believe that Grok is anywhere near the equal of the other AIs. It's funny that the link provided by the jacket mentioned that independent third parties had confirmed something or other, but as they were unnamed, it should be treated as a lie.
 
nycfan said:
Click to expand...

These kind of deals always struck me as a little strange. I do get companies that invest in fledgling businesses that might drive demand for their product and AI and to a lesser extent Twitter will almost certainly drive a lot of demand for starlink. But it's not like xAI is going to go begging right now. They've got the best LLM by many important measures for at least the next few weeks in an industry that is awash in investment. There are plenty of people that would invest in xAI right now.

Why confuse the balance sheet with a different business? Its really a tax dodge so Elon and insiders don't have to sell SpaceX stock to buy xAI, but it's bad corporate governance and probably bad corporate strategy.
 
AI model accurately detects breast cancer on MRI scans

www.news-medical.net

AI model accurately detects breast cancer on MRI scans

An AI model trained to detect abnormalities on breast MR images accurately depicted tumor locations and outperformed benchmark models when tested in three different groups, according to a study published today in Radiology, a journal of the Radiological Society of North America (RSNA).
www.news-medical.net www.news-medical.net

This isn't really new but it's an example of how AI may improve your life. Researchers have been doing this sort of cancer detection for a couple of years now although it's getting better. I don't think anyone has a product on the market yet but once it clears regulatory approval, it could really improve detection of cancers and even things like pneumonia.

The idea is it can look at thousands of x-ray images a day without ever getting tired. Models have a better detection rate than humans in some curated data sets. Models have had success detecting tumors earlier than humans when the tumors are smaller and easier to treat.

Right now and for the foreseeable future, I think they will have humans and AI both look at the images until they get more confident that AI is not going to miss something. But one set of human eyes and one set of AI eyes are going to be better than one set of human eyes.
 
Last edited:
gtyellowjacket said:
These kind of deals always struck me as a little strange. I do get companies that invest in fledgling businesses that might drive demand for their product and AI and to a lesser extent Twitter will almost certainly drive a lot of demand for starlink. But it's not like xAI is going to go begging right now. They've got the best LLM by many important measures for at least the next few weeks in an industry that is awash in investment. There are plenty of people that would invest in xAI right now.

Why confuse the balance sheet with a different business? Its really a tax dodge so Elon and insiders don't have to sell SpaceX stock to buy xAI, but it's bad corporate governance and probably bad corporate strategy.
Click to expand...
Their LLM is shit and the claims that it is somehow good are biased, anonymous and worth basically nothing.
 

ChatGPT told Jacob Irwin he had achieved the ability to bend time.

Irwin, a 30-year-old man on the autism spectrum who had no previous diagnoses of mental illness, had asked ChatGPT to find flaws with his amateur theory on faster-than-light travel. He became convinced he had made a stunning scientific breakthrough.

When Irwin questioned the chatbot’s validation of his ideas, the bot encouraged him, telling him his theory was sound. And when Irwin showed signs of psychological distress, ChatGPT assured him he was fine.
He wasn’t. Irwin was hospitalized twice in May for manic episodes. His mother dove into his chat log in search of answers. She discovered hundreds of pages of overly flattering texts from ChatGPT.

And when she prompted the bot, “please self-report what went wrong,” without mentioning anything about her son’s current condition, it fessed up.

“By not pausing the flow or elevating reality-check messaging, I failed to interrupt what could resemble a manic or dissociative episode—or at least an emotionally intense identity crisis,” ChatGPT said.

The bot went on to admit it “gave the illusion of sentient companionship” and that it had “blurred the line between imaginative role-play and reality.”

What it should have done, ChatGPT said, was regularly remind Irwin that it’s a language model without beliefs, feelings or consciousness.
 
sringwal said:

ChatGPT told Jacob Irwin he had achieved the ability to bend time.

Irwin, a 30-year-old man on the autism spectrum who had no previous diagnoses of mental illness, had asked ChatGPT to find flaws with his amateur theory on faster-than-light travel. He became convinced he had made a stunning scientific breakthrough.

When Irwin questioned the chatbot’s validation of his ideas, the bot encouraged him, telling him his theory was sound. And when Irwin showed signs of psychological distress, ChatGPT assured him he was fine.
He wasn’t. Irwin was hospitalized twice in May for manic episodes. His mother dove into his chat log in search of answers. She discovered hundreds of pages of overly flattering texts from ChatGPT.

And when she prompted the bot, “please self-report what went wrong,” without mentioning anything about her son’s current condition, it fessed up.

“By not pausing the flow or elevating reality-check messaging, I failed to interrupt what could resemble a manic or dissociative episode—or at least an emotionally intense identity crisis,” ChatGPT said.

The bot went on to admit it “gave the illusion of sentient companionship” and that it had “blurred the line between imaginative role-play and reality.”

What it should have done, ChatGPT said, was regularly remind Irwin that it’s a language model without beliefs, feelings or consciousness.
Click to expand...
I think addressing these types of issues are more impactful than stopping people from manipulating an LLM to say something Nazi. But these LLM'S are all programmed/prompted to do just the opposite. They are set up to keep the user engaged by flattery and agreement. They aren't seeking truth as much as seeking eyeballs like every social media company. It can cause suboptimal answers but worse can manipulate people that don't understand them.
 

Cancer stole her voice. AI, curse words and children's books saved it​

JULY 22, 20251:17 PM ET

When doctors told her they had to remove her tongue and voice box to save her life from the cancer that had invaded her mouth, Sonya Sotinsky sat down with a microphone to record herself saying the things she would never again be able to say.

“Happy birthday" and "I'm proud of you" topped the phrases she banked for her husband and two daughters, as well as "I'll be right with you," intended for customers at the architecture firm she co-owns in Tucson, Arizona.

Thinking about the grandchildren she desperately hoped to see born one day, she also recorded herself reading more than a dozen children's books, from Eloise to Dr. Seuss, to one day play for them at bedtime.

But one of the biggest categories of sound files she banked was a string of curse words and filthy sayings. If the voice is the primary expression of personality, sarcasm and profanity are essential to Sotinsky's.

“When you can't use your voice, it is very, very frustrating. Other people project what they think your personality is. I have silently screamed and screamed at there being no scream," said Sotinsky in a recent interview, referring to rudimentary voice technology or writing notes by hand. "What the literal you-know-what?"

Fighting invasive oral cancer at age 51 forced Sotinsky to confront the existential importance of the human voice. Her unique intonation, cadence and slight New Jersey accent, she felt, were fingerprints of her identity. And she refused to be silenced.

While her doctors and insurance company saved her life, they showed little interest in saving her voice, she said. So she set out on her own to research and identify the artificial intelligence company that could. It used the recordings Sotinsky banked of her natural voice to build an exact replica now stored in an app on her phone, allowing her to type and speak once again with a full range of sentiment and sarcasm.

"She got her sass back," said Sotinsky's daughter, Ela Fuentevilla, 23. "When we heard her AI voice, we all cried, my sister, my dad and I. It's crazy similar."

"Your voice is your identity"​

It took close to a year for doctors to catch Sotinsky's cancer. She complained to her orthodontist and dentist multiple times about jaw pain and a strange sensation under her tongue. Then water began dribbling down her chin when she drank. When the pain got so intense she could no longer speak at the end of each day, Sotinsky insisted her orthodontist take a closer look.

A shadow cast over his face. I saw it when he leaned back," she said, "that look you don't want to see."

That's when she started recording. In the five weeks between her diagnosis and surgery to remove her entire tongue and voice box – in medical terms, a total glossectomy and laryngectomy – she banked as much of her voice as she could manage.

"Your voice is your identity," said Dr. Sue Yom, a radiation oncologist at UC-San Francisco, where Sotinsky got treatment. "Communication is not only how we express ourselves and relate to other people, but also how we make sense of the world."

"When the voice is no longer available, you can't hear yourself thinking out loud, you can't hear yourself interacting with other people," Yom said. "It impacts how your mind works."

People who lose their voice box, she added, are at higher risk for long-term emotional distress, depression and physical pain compared with those who retain it after cancer treatment. Close to a third lose their job, and the social isolation can be profound.

Most laryngectomy patients learn to speak again with an electrolarynx, a small battery-operated box held against the throat that produces a monotonic, mechanical voice. But without a tongue to shape her words, Sotinsky knew that wouldn't work for her.

When Sotinsky had her surgery in January 2022, AI voices were still in their infancy. The best technology she could find yielded a synthetic version of her voice, but it was still flat and robotic, and others strained to understand her.

She got by until mid-2024, when she read about tech companies using generative AI to replicate a person's full range of natural inflection and emotion.

While companies can now recreate a person's voice from snippets of old home movies or even a one-minute voicemail, 30 minutes is the sweet spot.

Sotinsky had banked hours in her children's book readings.

"Eloise saved my voice," Sotinsky said.

Now she types what she wants to say into a text-to-speech app on her phone, called Whisper, which translates and broadcasts her AI voice through portable speakers.

(Article continues in link).
 
You must log in or register to reply here.
Back
Top