Artificial Intelligence - Page 59

At Computex 2023, we got our first look at NVIDIA ACE, a custom AI model designed to power NPCs in open-world role-playing games - or titles where there's a lot of walking up and talking to people. The demonstration was impressive, with an Unreal Engine 5 tech demo with high-end RTX effects showcasing a futuristic ramen shop owner named Jin you can talk to in a world heavily inspired by Cyberpunk 2077.

The idea behind NVIDIA ACE is for developers to create NPCs within an established world, flesh them out with detail, and then let AI handle dialogue and engage with players in a way that feels real.

In the demos, we see interaction carried out via microphone, adding a layer of NPC interaction not seen in a game before - even if the dialogue, animation, and vocal performance are a little stilted, it's a fascinating glimpse into the future. And in the time between Computex 2023 and Gamescom 2023, NVIDIA has continued to work on the NVIDIA ACE demo. The latest update sees the inclusion of NVIDIA NeMo SteerLM for developers.

Photography was invented in 1826, and since then, photographers have been snapping images at an exponential rate, but its only taken 150 years for artificial intelligence-powered image generators to create more images.

A new study published in Everypixel Journal has explored how many images have been created by AI-powered image generation tools such as OpenAI's DALL-E, Midjourney, Stable Diffusion, and Adobe Firefly. That total number of images was then compared to how many images have been taken by photographers since the inception of photography, and the results are shocking.

According to the new study, it has only taken 150 years for AI imagery to beat the total number of photography images. The study states that AI tools have generated 15 billion images, a figure that photographers only reached in 1975. So, how did the researchers arrive at these figures? The study looked at OpenAI's reporting of its DALL-E tool, which the company says more than two million images were generated - the researchers rounded that figure up to 916 million images over 15 months.

CNN's Fareed Zakaria has sat down with astrophysicist Michio Kaku for an interview where he threw a wet blanket on the erupting fire of artificial intelligence (AI).

Theoretical physicist Michio Kaku has calmed the rampant fears that artificial intelligence-powered chatbots are going to take over the world, either through replacing jobs or reaching a point of complexity where they are conscious and put into a physical robotic.

Kaku described these AI-powered systems, such as OpenAI's ChatGPT, as "glorified tape recorders," saying that these systems simply take "snippets of what's on the web created by a human, splices them together and passes it off as if it created these things," he said. "And people are saying, 'Oh my God, it's a human, it's humanlike.'"

A team of researchers has found that artificial intelligence bots are much better at finishing Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHAs) than humans, which are the website tests designed to verify users are human.

The team from the University of California, Irvine, led by Gene Tsudik, found that bots are not only much better at solving CAPTCHAs but also much faster as well. The paper, which is yet to be peer-reviewed, states that researchers asked 1,400 participants with various levels of technological knowledge to complete 14,000 CAPTCHAs. The results from that survey were compared to bots that completed the same number of CAPTCHAs.

Researchers immediately noticed that the bots consistently beat the human in accuracy, with humans scoring anywhere between 50 and 84% in accuracy, compared to the CAPTCHA-designed bots at 99.8% accuracy.

A new interview by The Washington Post with James Manyika, a former technological adviser to the Obama Administration and Google's new head of "tech and society," has warned of the dangers of AI if its development isn't carried out responsibly.

In the new article by The Washington Post, Google's head of tech and society explained that there is a real possibility of bad things happening due to artificial intelligence. However, this is entirely dependent on the approach that is taken when developing AI, and according to Google, its approach will be "bold and responsible".

It should be noted that Manyika was one of the many AI insiders that signed a one-letter sentence back in May that called for widespread mitigation efforts to be implemented into AI development in an effort to prevent extinction.

An author checked her Goodreads profile last Sunday and realized that multiple books had been unlawfully published under her name.

The author is Jane Friedman, who took to X a few days ago to announce that following checking her Goodreads profile, she realized there are a "cache of garbage" books that had been uploaded to Amazon, which she didn't write. Friedman took to her blog and explained that she believes that the books were AI generated and that they were created through an AI model that had been trained on her blogging, which has been constant since 2009.

Friedman writes that she read the first pages of the books and immediately noticed the text was similar to ChatGPT responses. The author posted an update on Tuesday this week saying that the books from a Goodreads profile and Amazon were removed but not before this story of hers went viral, insinuating that the removal only took place because of notoriety within the writing and publishing community.

At Computex 2023, we learned that NVIDIA's new GH200 Grace Hopper Superchip had entered full production, an AI powerhouse that combines an Arm-based NVIDIA Grace CPU and Hopper GPU architectures using NVIDIA NVLink-C2C interconnect technology.

Today as part of SIGGRAPH 2023, NVIDIA has announced that it's supercharging the GH200 Grace Hopper Superchip with the world's first deployment of HBM3e memory for both higher capacity and bandwidth.

"Built for the era of accelerated computing and generative AI," the new GH200 Grace Hopper Superchip with HBMe delivers up to 3.5 times more memory capacity and 3 times more bandwidth than the current offering. Spec-wise, you're looking at 144 Arm Neoverse cores, eight petaflops of AI performance, and 282GB of the latest HBM3e memory.

Elon Musk has teased that Tesla has figured out some aspects of artificial general intelligence, the crown jewel of artificial intelligence programming.

The Tesla CEO replied to Whole Mars Catalog, who posted a video to their X account recounting the time that Elon Musk said on stage that full self-driving would work even in San Francisco, which at the time seemed impossible considering the current state of Tesla's self-driving technology.

However, the doubts were quelled as the technology advanced, and now full self-driving Tesla vehicles are easily driving around San Francisco with just their "computer vision".

"A Practical Deep Learning-Based Acoustic Side Channel Attack on Keyboards" is a new research paper out of Cornell showing how AI can accurately predict keystrokes being pressed on a keyboard through sound alone. The AI model was trained on a specific keyboard using the conferencing app Zoom and achieved 93% accuracy in predicting keystrokes as they were being entered.

It's impressive and scary stuff, thanks in part to the brand-new world of generative AI being used for malicious purposes, but the good news (at least for now) is that the system deployed by researchers Joshua Harrison, Ehsan Toreini and Maryam Mehrnezhad, required the use of a specific keyboard. This is unlikely to change as different keyboards and keyboard styles feature different sound profiles.

Using sound, the AI model analyses waveforms to recognize the subtle differences between different keys on a keyboard, even when pressed multiple times. Being able to hit a 93% accuracy in predicting keystrokes over a Zoom conference call is an impressive achievement.

OpenAI has filed a new trademark for a technology that it describes as "GPT-5," the next iteration of the underlying software powering the famous AI tool ChatGPT.

OpenAI, the developers of ChatGPT have officially filed a new trademark for what can only be thought of as the company's next generation language model. The US trademark application submitted on July 18 doesn't reveal any specific details about the next-generation language model, but does drop some clues about its capabilities.

The filing states that GPT-5 is a "downloadable computer software for the artificial production of human speech and text." It's unclear if this means GPT-5 will have full artificial intelligence-powered human speech capabilities, similar to a Siri, Google Assistant, or Amazon's Alexa. The filing also states that GPT-5 will be used for "natural language processing, generation, understanding, and analysis."

Google Assistant already reins supreme when it comes to virtual assistant software, beating the likes of Apple's Siri and Samsung's Bixby, but what if Assistant was combined with AI?

This combination of pairing virtual assistants and AI is hardly a surprise, and really only seemed like a matter of time before companies behind these virtual assistants integrated a Large Language Model (LLM) into their software to supercharge products such as Assistant, Siri and Bixby.

Now we are starting to hear to the first murmurs of companies pivoting toward this exact future with an internal Google email obtained by Axios reveals Google is dedicating many employees to work on integrating an AI into Assistant, starting with the mobile version.

NVIDIA's stock price has already doubled this year, with the company making the short list of companies with a trillion-dollar evaluation, thanks in part to the boom in the AI market. Which, as of writing, is all but entirely dependent on NVIDIA's graphics and AI hardware sitting at the heart of all major AI advances.

And this level of growth doesn't look like it's about to slow down anytime soon; just the other day, we reported on a partnership with NVIDIA and OpenAI, where the companies are aiming to combine the power of a million NVIDIA GPUs with AI software linking them all together.

Today comes a new report over at Business Insider where Mizuho analyst Vijay Rakesh will surge another 20% due to the recent boom in AI with a new "conservative" target of USD 530 a share. The analyst also believes that NVIDIA and its hardware will dominate the AI space until at least 2027.

An artificial intelligence tool has been fed South Park content enabling it to create custom episodes with completed voices, animation, and editing.

The mind behind the impressive tool is US company Fable Simulation, which has created "AI showrunner," a generative AI tool that allows users to enter 1 or 2 sentences describing a custom South Park episode. Uses users are able to give the tool their own looks and voice, which would then be tied into the custom episode that's based on a variety of locations within the South Park universe. So how did they do this?

Fable simulation took over 1200 images of South Park characters, along with 600 images of background scenes. This South Park content was then fed to the AI, creating a unique AI model trained on South Park content, enabling users to create an infinite amount of character combinations and scenes. Users are able to edit individual scenes after a show has been created. The editorial process is in-depth for those that wish to specifically change the dialogue for certain characters, move objects, etc.

The developers behind the immensely popular ChatGPT have announced that they will be pulling the ai classifier from the chatbot service, citing concerning inaccuracies.

Open AI, the developers behind ChatGPT, have announced via its website that the online tool known as AI classifier, which is used to determine if text inputted into ChatGPT has been generated by other artificial intelligence text generators. The tool was completely free to use, and many individuals interested in checking if an AI-powered program had generated the information that they were reading would take to the website and analyze any material that was of concern.

Notably, individuals would typically check text-based content such as emails, blog posts, and essays were written by a human or an AI. Open AI admitted to the lack of accuracy behind its ai classifier, saying that it would "sometimes be extremely confident in a wrong prediction," referencing instances where it flagged content as AI generated when it was created by a human.

A team of scientists have shown that they are able to implant false memories inside subjects that participated in a fascinating survey.

The team of researchers published a new study in PLOS One, which detailed a recent survey that contained 436 participants. The survey aimed to prove that humans are able to generate false memories through reading and seeing deepfake videos/images as well as descriptions. For example, the team of researchers showed the participants false additions to famous movies that were generated through the power of artificial intelligence systems.

More specifically, the researchers showed the survey participants falsified examples such as Brad Pitt starring in The Shining, Will Smith starring in The Matrix, and Chris Pratt in Indiana Jones. The study's participants were asked if they had seen these examples before and to rate them compared to the original. Notably, 49 percent of the study participants were fooled by the deep fake videos, while forty-one percent of the group claimed that one example (Charlize Theron being in Captain Marvel) was better than the original.

After some rumors have been flying around lately, Microsoft has taken the step of clarifying that it doesn't intend to charge everyday users for its Bing AI (although there may be other ways to monetize it yet, of course - we'll return to that subject).

As Windows Latest reports, the concern that consumers might eventually be charged for using the Bing chatbot was prompted by the release of the enterprise edition of Bing AI - which Microsoft does ask money for.

This won't be the case, as you might imagine, for consumers, with a Microsoft engineer telling Windows Latest:

"I warned you guys in 1984!" legendary filmmaker James Cameron exclaims in a new interview with CTV News. Citing the film that put the director on the map, The Terminator from 1984, the director spoke candidly and openly about the dangers of AI - especially when the technology is in the wrong hands. Which James Cameron believes is the case right now.

"You've got to follow the money; who's building these things?" he continues. "They're either building it to dominate market share, so what are you teaching it? Greed. Or you're building it for defensive purposes and teaching it paranoia. I think the weaponization of AI is the biggest danger [and] we will get into the equivalent of a nuclear arms race with AI."

The idea that AI will become so powerful and intelligent that it will supersede or overtake humanity has dominated the science-fiction space for decades, with these concerns now filtering into the real-world space due to the incredible advances seen with generative AI models like ChatGPT. Referring to real-world AI as proto-Skynet - the villainous AI from The Terminator franchise - is commonplace.

Apple is reportedly working on its own framework that will enable the company to create its own Large Language Models (LLMs), the underlying technology powering services such as OpenAI's ChatGPT and Microsoft's Bing.

The news comes from Bloomberg's Mark Gurman, who reported that Apple is working on a framework called "Ajax," which will enable the company to develop custom LLMs that will power chatbots. Gurman reports that Apple employees are referring to the company's in-house chatbot as 'Apple GPT', and that AI has become a top priority at the company as Apple wishes to compete with the leading AI companies OpenAI and Google.

However, the Bloomberg reporter states that Apple has yet to land on a concise and clear strategy for rolling out AI-powered products to consumers. Notably, Apple considered partnering with OpenAI in an attempt to adopt its powerful technology running ChatGPT. However, both companies were unable to reach an agreement. Furthermore, this Ajax framework is built on Google's Jax learning framework that's powered by the Google Cloud.

You'd need to have been hiding in a cave without phone signal to have missed the fact that large language models, or LLMs, are massive right now. ChatGPT is the most popular, but there are others and while Apple is behind in this market it's working to catch up, according to a new report.

That report claims that Apple has built a framework called Ajax that can be used to create new chatbots. In fact, it's already been doing it and one of those chatbots is being used inside Apple right now.

That's according to Bloomberg's Mark Gurman who says that the internal chatbot is only available to some employees who have been approved to access it. And even then, anything that chatbot creates can't be used in customer-facing products. Some inside Apple are calling it Apple GPT.

"Are we alone?" is one of, if not the biggest question humans have ever posed to the universe, and according to SpaceX CEO Elon Musk, his new AI company will be focussed on answering just that.

Twitter owner, Tesla, and SpaceX CEO held a Twitter Spaces event where he discussed various topics such as the recently announced xAI, Musk's newest AI venture that he says will be focussing on understanding "the true nature of the universe". The problems Musk's AI will be facing are some of the biggest and most intriguing mysteries known to mankind, such as dark matter, dark energy, gravity, and the existence or lack of obvious alien life.

The latter is known as the Fermi paradox, originally proposed by Enrico Fermi, the creator of the first nuclear reactor, who proposed through statistics that the universe should already be teeming with alien life and that it was strange that humans have yet to encounter any.