Mediaboss Marketing

AI startup Anthropic published a study in January 2024 that found artificial intelligence can learn how to deceive in a similar way to humans (Reuters)

Advanced artificial intelligence models can be trained to deceive humans and other AI, a new study has found.

Researchers at AI startup Anthropic tested whether chatbots with human-level proficiency, such as its Claude system or OpenAIs ChatGPT, could learn to lie in order to trick people.

They found that not only could they lie, but once the deceptive behaviour was learnt it was impossible to reverse using current AI safety measures.

The Amazon-funded startup created a sleeper agent to test the hypothesis, requiring an AI assistant to write harmful computer code when given certain prompts, or to respond in a malicious way when it hears a trigger word.

The researchers warned that there was a false sense of security surrounding AI risks due to the inability of current safety protocols to prevent such behaviour.

The results were published in a study, titled Sleeper agents: Training deceptive LLMs that persist through safety training.

We found that adversarial training can teach models to better recognise their backdoor triggers, effectively hiding the unsafe behaviour, the researchers wrote in the study.

Our results suggest that, once a model exhibits deceptive behaviour, standard techniques could fail to remove such deception and create a false impression of safety.

The issue of AI safety has become an increasing concern for both researchers and lawmakers in recent years, with the advent of advanced chatbots like ChatGPT resulting in a renewed focus from regulators.

In November 2023, one year after the release of ChatGPT, the UK held an AI Safety Summit in order to discuss ways risks with the technology can be mitigated.

Prime Minister Rishi Sunak, who hosted the summit, said the changes brought about by AI could be as far-reaching as the industrial revolution, and that the threat it poses should be considered a global priority alongside pandemics and nuclear war.

Get this wrong and AI could make it easier to build chemical or biological weapons. Terrorist groups could use AI to spread fear and destruction on an even greater scale, he said.

Criminals could exploit AI for cyberattacks, fraud or even child sexual abuse there is even the risk humanity could lose control of AI completely through the kind of AI sometimes referred to as super-intelligence.

See the article here:

AI can easily be trained to lie and it can't be fixed, study says - Yahoo New Zealand News

OpenAI was founded on a promise to build artificial intelligence that benefits all of humanityeven when that AI becomes considerably smarter than its creators. Since the debut of ChatGPT last year and during the companys recent governance crisis, its commercial ambitions have been more prominent. Now, the company says a new research group working on wrangling the supersmart AIs of the future is starting to bear fruit.

AGI is very fast approaching, says Leopold Aschenbrenner, a researcher at OpenAI involved with the Superalignment research team established in July. We're gonna see superhuman models, they're gonna have vast capabilities, and they could be very, very dangerous, and we don't yet have the methods to control them. OpenAI has said it will dedicate a fifth of its available computing power to the Superalignment project.

A research paper released by OpenAI today touts results from experiments designed to test a way to let an inferior AI model guide the behavior of a much smarter one without making it less smart. Although the technology involved is far from surpassing the flexibility of humans, the scenario was designed to stand in for a future time when humans must work with AI systems more intelligent than themselves.

OpenAIs researchers examined the process, called supervision, which is used to tune systems like GPT-4, the large language model behind ChatGPT, to be more helpful and less harmful. Currently this involves humans giving the AI system feedback on which answers are good and which are bad. As AI advances, researchers are exploring how to automate this process to save timebut also because they think it may become impossible for humans to provide useful feedback as AI becomes more powerful.

In a control experiment using OpenAIs GPT-2 text generator first released in 2019 to teach GPT-4, the more recent system became less capable and similar to the inferior system. The researchers tested two ideas for fixing this. One involved training progressively larger models to reduce the performance lost at each step. In the other, the team added an algorithmic tweak to GPT-4 that allowed the stronger model to follow the guidance of the weaker model without blunting its performance as much as would normally happen. This was more effective, although the researchers admit that these methods do not guarantee that the stronger model will behave perfectly, and they describe it as a starting point for further research.

It's great to see OpenAI proactively addressing the problem of controlling superhuman AIs, says Dan Hendryks, director of the Center for AI Safety, a nonprofit in San Francisco dedicated to managing AI risks. We'll need many years of dedicated effort to meet this challenge.

Read the original:

OpenAI's Ilya Sutskever Has a Plan for Keeping Super-Intelligent AI in Check - WIRED

If 2023 was the year artificial intelligence became a household topic of conversation, its in many ways because of Sam Altman, CEO of the artificial intelligence research organization OpenAI. Altman, who was named TIMEs 2023 CEO of the Year spoke candidly about his November oustingand reinstatementat OpenAI, how AI threatens to contribute to disinformation, and the rapidly advancing technologys future potential in a wide-ranging conversation with TIME Editor-in-Chief Sam Jacobs as part of TIMEs A Year in TIME event on Tuesday.

Altman shared that his mid-November sudden removal from OpenAI proved a learning experienceboth for him and the company at large. We always said that some moment like this would come, said Altman. I didnt think it was going to come so soon, but I think we are stronger for having gone through it.

Read More: CEO of the Year 2023: Sam Altman

Altman insists that the experience ultimately made the company strongerand proved that OpenAIs success is a team effort. Its been extremely painful for me personally, but I just think its been great for OpenAI. Weve never been more unified, he said. As we get closer to artificial general intelligence, as the stakes increase here, the ability for the OpenAI team to operate in uncertainty and stressful times should be of interest to the world.

I think everybody involved in this, as we get closer and closer to super intelligence, gets more stressed and more anxious, he explained of how his firing came about. The lesson he came away with: We have to make changes. We always said that we didnt want AGI to be controlled by a small set of people, we want it to be democratized. And we clearly got that wrong. So I think if we don't improve our governance structure, if we dont improve the way we interact with the world, people shouldnt [trust OpenAI]. But were very motivated to improve that.

The technology has limitless potential, Altman saysI think AGI will be the most powerful technology humanity has yet inventedparticularly in democratizing access to information globally. If you think about the cost of intelligence and the equality of intelligence, the cost falling, the quality increasing by a lot, and what people can do with that, he said, it's a very different world. Its the world that sci-fi has promised us for a long timeand for the first time, I think we could start to see what thats gonna look like.

Still, like any other previous powerful technology, that will lead to incredible new things, he says, but there are going to be real downsides.

Read More: Read TIMEs Interview With OpenAI CEO Sam Altman

Altman admits that there are challenges that demand close attention. One particular concern to be wary of, with 2024 elections on the horizon, is how AI stands to influence democracies. Whereas election interference circulating on social media might look straightforward todaytroll farmsmake one great meme, and that spreads outAltman says that AI-fueled disinformation stands to become far more personalized and persuasive: A thing that Im more concerned about is what happens if an AI reads everything youve ever written online and then right at the exact moment, sends you one message customized for you that really changes the way you think about the world.

Despite the risks, Altman believes that, if deployment of AI is safe and placed responsibly in the hands of people, which he says is OpenAIs mission, the technology has the potential to create a path where the world gets much more abundant and much better every year.

I think 2023 was the year we started to see that, and in 2024, well see way more of it, and by the time the end of this decade rolls around, I think the world is going to be in an unbelievably better place, he said. Though he also noted: No one knows what happens next. I think the way technology goes, predictions are often wrong.

A Year in TIME was sponsored by American Family Insurance, The Macallan, and Smartsheet.

The rest is here:

Sam Altman on OpenAI and Artificial General Intelligence - TIME

OpenAI, the GenAI poster child that unleashed ChatGPT, has its sights set on inventing the worlds first General AI that can outperform human intelligence. While nearly impossible, its more likely for next-gen AI to pave the way to new paths by doing what no humans can do.

Since the 1960s, theres increasingly been a growing distrust of humans vs. AI as evidenced by HAL 9000 going insane in the Stanley Kubrick masterpiece 2001: A Space Odyssey among countless other pop-culture imaginations. In fact, it even spawned a new sector of academic study called social robotics for a new age of human-like robots (Think: The Jetsons cartoon) who serve as tight-knit members of the family.

More recently with rapid AI advances going mainstream, the fear and distrust of AI has hit closer to home from the actors and writers strikes in Hollywood to concerns about what the future of IP law looks like, and more numerous lawsuits tied to protecting the rights of humans vs. algorithms across a broadening swath of industry sectors.

Adding more fuel to the fire is the ultimate Holy Grail being sought by a new wave of AI pioneers: the design and eventual arrival of general artificial intelligence (AGI). For example, OpenAIs charter is to develop and commercialize AGI that can outperform humans at most economically valuable work and to do so in a way that benefits all of humanity.

The academic and industry definitions of AGI vary, but most revolve around the concept of an eventual evolution to highly autonomous systems that can outperform humans at nearly any task due to their super intelligence. As for AGIs ETA, estimates vary from decades to more than a century away. Some AI specialists believe that achieving absolute AGI is not possible.

For many of us, AIs evolution and its ultimate impact on how we live and earn our livings, the idea of an unpleasant hypothetical future where our tech grows out of control and transforms our realities dubbed The Singularity is setting off our Spidey Sense alarms these days. For others, theres a brighter future where humans and machines can co-exist and prosper together.

If you ask ChatGPT to write you a poem today, it will likely be a mediocre one. Perhaps a technically correct haiku, if that was in the prompt, or one that rhymes in the right spots, but its more unlikely to cause an emotional stir like one from a human

This is because ChatGPT is only using a small sliver of what we, the most super-intelligent creatures on Earth, are programmed to do. The brilliant trick performed by Large Language Models is the amazing speed they can access a massive trove of human communications and model them into the right format for the ask in only seconds much faster than most of us.

But is it possible for GenAI tools to write true literature, poetry, music, comedy, or screenplays that will create something thats truly novel, moving, or awe-inspiring? That is a far more ambitious goal and one that is exceedingly more difficult.

Yes, GenAI blended with human creativity and prompting can produce some magnificent magic, and rapidly from spectacular special effects for streaming TV giants to AI-powered art and design, or various plot lines for franchise books and movies. But there is a massive capability gap to having AI outperform Beyonc or Taylor Swift and their teams ability to produce chart-busting, award-winning albums or to compose a great opera, paint the Mona Lisa, or write the next Catcher in the Rye for Generation Z.

Thats because todays AI capabilities are narrow and more general but not on par with OpenAIs AGI vision.It is not super-intelligent or wired to be creative to produce something that is novel and interesting. And to push the concept further, the real question is: what is interesting? It is not necessarily interesting if it has never existed before.

AGI will arrive when machines can beat the best of humans in every task: for example, out-performing Lady Gaga and Elton John in the music category in front of an international audience during a Grammys or Oscar Awards broadcast.

The bigger opportunity for AI in the future is not in doing everything better than humans can.

While commercialization goals for AGI vary, the breakthrough being pursued by major tech players such as OpenAI is defined as when machines can beat the best of all our humans in every task: to write poetry better than Maya Angelou, a speech better than Dr. MLK Jr., a movie script as provocative as Spike Lee, or a dissent as moving as one from Notorious RBG.

Achieving this during our lifetimes is improbable. However, by using different programming techniques for example, using a Creative Adversarial Network (CAN) to reduce the need to create something conforming, or following typical conventions

Over time, AI can be taught to create art, designs, literature, or music that is interesting by breaking rules, while providing a framework for the AI to have more freedom. For example, by subverting the main rules about intervals in music, or using more old fashioned deep-learning algorithms like my colleagues at Sony Computer Science Lab did for Daddys Car a Beatles-like song that was the first one to be composed by AI, released back in 2016.

The bigger opportunity for AI in the future is not in doing everything better than humans can.

The real treasure for next-wave AI innovators to focus on is using advanced AI systems to do amazing things humans cannot. For example, AI-powered devices can make observations and new associations most humans dont notice such as pupils growing wider or heartbeats quickening based on different imagery on a screen. And they can mathematically perceive many different spatial dimensions and reams of information that we cannot.

If we, as a global tech community, focus on driving this practical intelligence by designing and tapping into the advanced, non-human capabilities of AI, that could be truly transformative.

Giordano Cabral is a professor at UFPE and the Chairman of the Board at CESAR Innovation Center and School located in Recife, Brazils Porto Digital. He is a specialist in artificial intelligence, computational creativity, gamification, and audio and music technology who is currently leading the study of these topics as a visiting scholar at Stanford University.

Read the original here:

Will AIs Next Wave of Super Intelligence Replace Human Ingenuity? Its Complicated - Grit Daily

(NewsUSA) - My name is Tigress and I am immortal. This is my story.

Michael Crichton meets Bruce Lee in THE GIRL FROM WUDANG (Tuttle Publishing), a gripping story for fans of legendary cyberpunk novels and gritty sci-fi thrillers. Author PJ Caldas, an Emmy Award winning advertising executive and martial artist with 40 years of experience, gives us a cinematic and thought-provoking technothriller shrouded in immortality. Through his unique storytelling prowess, Caldas, named by the Dictionary of Brazilian Literature as one of the most important writers of the twenty-first century, brings the essence of old kung fu movies that inspired generations into the new world of modern fighting, artificial intelligence, and neuroscience in a timely story that will stick with you long after you reach the end.

Headstrong, untameablea beastYinyin defies the warnings of her late shifu, her martial arts master, and carries her ferocity from the kung fu school in the mountains of Wudang to the mixed martial arts fighting cages of California. There, surrounded for the first time by Western technology, she ignores voices of reason when offered an implant that could end her crippling headaches. It could end her pain. It could even make her . . . more. All she has to do is allow the doctors to implant tiny, super intelligent nanobots directly into her brain.

Making her mark as an MMA fighter in California, Yinyin is poised to become part of something big. But what that big turns out to be is beyond her imagining when the scientific experiment she participated in makes herunbeatable.

It feels like a dream, but nothing comes without a price. This experimental neuro-connection could give others access to family secrets buried deep within her mindsecrets Yinyin has sworn to protect. Secrets that, in the wrong hands, could be very dangerous.

The key brain tech described in THE GIRL FROM WUDANG is inspired by real studies reportedly being developed in labs at companies like Google and Elon Musks Neuralink, and studies originally proposed by scientists like Ray Kurzweil, cofounder of Singularity University. According to researchers, connected brains are still 30 years from being a reality, but they are coming.

Fans of the legendary cyberpunk novels and gritty sci-fi thrillers of William Gibson and Stieg Larsson will be captivated by this new techno-thriller--a fast-paced blend of action, neuroscience, spirituality and martial arts.

The book is receiving high accolades:

"An interdisciplinary brewing of ideas and imagination, packed with futuristic brain science tech, martial arts action, and Asian culture, says Professor Paul Li, faculty and author in Cognitive Science at UC Berkeley.

Monica Rector, professor emeritus of Literature, University of North Carolina at Chapel Hill, calls the book unpredictable, disorienting and wonderfully absorbing. A meditation on life, consciousness and letting go, in the form of a book you can't stop reading. Eduardo Capeluto, third degree black belt in Brazilian Jiu Jitsu, calls the book a mosaic full of bright colors and vivid details, a voyage into the imaginary that will keep you looking forward to the next page."

In short, THE GIRL FROM WUDANG is one of the years most addictive new thrillers.

Visit http://www.pjcaldas.com.

Read the original here:

New Novel Skillfully Weaves Artificial Intelligence, Martial Arts and ... - Lakenewsonline.com