Cloud Hosting

A new artificial intelligence (AI) tool can generate images in under two seconds and it doesn't need expensive hardware to run.

South Korean scientists have used a special technique called knowledge distillation to compress the size of an open source (or publicly available) image generation model known as Stable Diffusion XL which has 2.56 billion parameters, or variables the AI uses to learn during training.

The smallest version of the new model, known as "KOALA", has just 700 million parameters, meaning it's lean enough to run quickly and without needing expensive and energy-intensive hardware.

Related: AI chatbots need to be much better at remembering things. Have scientists just cracked their terrible memory problem?

The method they used, knowledge distillation, transfers knowledge from a large model to a smaller one, ideally without compromising performance. The benefit of a smaller model is that it takes less time to perform computations and generate an answer.

The tool can run on low-cost graphics processing units (GPUs) and needs roughly 8GB of RAM to process requests versus larger models, which need high-end industrial GPUs.

The team published their findings in a paper Dec. 7, 2023 to the preprint database arXiv. They have also made their work available via the open source AI repository Hugging Face.

The Electronics and Telecommunication Research Institute (ETRI), the institution behind the new models, has created five versions including three versions of the "KOALA" image generator which generates images based on text input and two versions of "Ko-LLaVA" which can answer text-based questions with images or video.

When they tested KOALA, it generated an image based on the prompt "a picture of an astronaut reading a book under the moon on Mars" in 1.6 seconds. OpenAI's DALLE 2 generated an image based on the same prompt in 12.3 seconds, and DALLE 3 generated it in 13.7 seconds, according to a statement.

The scientists now plan to integrate the technology they've developed into existing image generation services, education services, content production and other lines of business.

Read more:

New AI image generator is 8 times faster than OpenAI's best tool and can run on cheap computers - Livescience.com

Alibaba wants you to compare its new AI video generator to OpenAI's Sora. Otherwise, why use it to make Sora's most famous creation belt out a Dua Lipa song?

On Tuesday, an organization called the "Institute for Intelligent Computing" within the Chinese e-commerce juggernaut Alibaba released a paper about an intriguing new AI video generator it has developed that's shockingly good at turning still images of faces into passable actors and charismatic singers. The system is called EMO, a fun backronym supposedly drawn from the words "Emotive Portrait Alive" (though, in that case, why is it not called "EPO"?).

EMO is a peek into a future where a system like Sora makes video worlds, and rather than being populated by attractive mute people just kinda looking at each other, the "actors" in these AI creations say stuff or even sing.

Alibaba put demo videos on GitHub to show off its new video-generating framework. These include a video of the Sora lady famous for walking around AI-generated Tokyo just after a rainstorm singing "Don't Start Now" by Dua Lipa and getting pretty funky with it.

The demos also reveal how EMO can, to cite one example, make Audrey Hepburn speak the audio from a viral clip of Riverdale's Lili Reinhart talking about how much she loves crying. In that clip, Hepburn's head maintains a rather soldier-like upright position, but her whole face not just her mouth really does seem to emote the words in the audio.

In contrast to this uncanny version of Hepburn, Reinhart in the original clip moves her head a whole lot, and she also emotes quite differently, so EMO doesn't seem to be a riff on the sort of AI face-swapping that went viral back in the mid-2010s and led to the rise of deepfakes in 2017.

Over the past few years, applications designed to generate facial animation from audio have cropped up, but they haven't been all that inspiring. For instance, the NVIDIA Omniverse software package touts an app with an audio-to-facial-animation framework called "Audio2Face" which relies on 3D animation for its outputs rather than simply generating photorealistic video like EMO.

Despite Audio2Face only being two years old, the EMO demo makes it look like an antique. In a video that purports to show off its ability to mimic emotions while talking, the 3D face it depicts looks more like a puppet in a facial expression mask, while EMO's characters seem to express the shades of complex emotion that come across in each audio clip.

It's worth noting at this point that, like with Sora, we're assessing this AI framework based on a demo provided by its creators, and we don't actually have our hands on a usable version that we can test. So it's tough to imagine that right out of the gate this piece of software can churn out such convincingly human facial performances based on audio without significant trial and error, or task-specific fine-tuning.

The characters in the demos mostly aren't expressing speech that calls for extreme emotions faces screwed up in rage, or melting down in tears, for instance so it remains to be seen how EMO would handle heavy emotion with audio alone as its guide. What's more, despite being made in China, it's depicted as a total polyglot, capable of picking up on the phonics of English and Korean, and making the faces form the appropriate phonemes with decent though far from perfect fidelity. So in other words, it would be nice to see what would happen if you put audio of a very angry person speaking a lesser-known language into EMO to see how well it performed.

Also fascinating are the little embellishments between phrases pursed lips or a downward glance that insert emotion into the pauses rather than just the times when the lips are moving. These are examples of how a real human face emotes, and it's tantalizing to see EMO get them so right, even in such a limited demo.

According to the paper, EMO's model relies on a large dataset of audio and video (once again: from where?) to give it the reference points necessary to emote so realistically. And its diffusion-based approach apparently doesn't involve an intermediate step in which 3D models do part of the work. A reference-attention mechanism and a separate audio-attention mechanism are paired by EMO's model to provide animated characters whose facial animations match what comes across in the audio while remaining true to the facial characteristics of the provided base image.

It's an impressive collection of demos, and after watching them it's impossible not to imagine what's coming next. But if you make your money as an actor, try not to imagine too hard, because things get pretty disturbing pretty quick.

Original post:

Alibaba's AI video generator just dunked on Sora by making the Sora lady sing - Mashable

Elon Musk has sued OpenAI, its co-founders Sam Altman and Greg Brockman, and its affiliated entities, alleging the ChatGPT makers have breached their original contractual agreements by pursuing profits instead of the nonprofits founding mission to develop AI that benefits humanity.

Musk, a co-founder and early backer of OpenAI, claims Altman and Brockman convinced him to help found and bankroll the startup in 2015 with promises it would be a nonprofit focused on countering the competitive threat from Google. The founding agreement required OpenAI to make its technology freely available to the public, the lawsuit alleges.

The lawsuit, filed in a court in San Francisco late Thursday, says that OpenAI, the worlds most valuable AI startup, has shifted to a for-profit model focused on commercializing its AGI research after partnering with Microsoft, the worlds most valuable company, which has invested about $13 billion into the startup.

In reality, however, OpenAI, Inc. has been transformed into a closed-source de facto subsidiary of the largest technology company in the world: Microsoft. Under its new board, it is not just developing but is actually refining an AGI to maximize profits for Microsoft, rather than for the benefit of humanity, the lawsuit adds. This was a stark betrayal of the Founding Agreement.

The lawsuit follows Musk airing concerns about OpenAIs shift in priorities in the past year. According to the legal complaint, Musk donated over $44 million to the nonprofit between 2016 and September 2020. For the first several years, he was the largest contributor to OpenAI, the lawsuit adds. Musk, who left OpenAIs board in 2018, has been offered a stake in the for-profit arm of the startup but has refused to accept it over a principled stand, he said earlier.

X, the social network owned by Musk, last year launched Grok, a rival to ChatGPT.

Altman has also addressed some of Musks concerns in the past, including the close ties with Microsoft. I like the dude. I think hes totally wrong about this stuff, he said of Musks criticisms at a conference last year. He can sort of say whatever he wants but Im like proud of what were doing and I think were going to make a positive contribution to the world and I try to stay above all that.

OpenAIs launch of ChatGPT in late 2022sparked an AI arms race, with rivals still scrambling to match its uncannily human-like responses. Microsoft CEO Satya Nadella landed a gloved jab at the rest of the industry last month. We have the best model today even with all the hoopla, one year after, GPT4 is better, he said. We are waiting for the competition to arrive. It will arrive, Im sure, but the fact [is] that we have the leading LLM out there.

An email exchange between Musk and Altman, presented as evidence in the lawsuit. Image Credits: TechCrunch/screenshot

The Thursday lawsuit alleges close alignment between Microsoft and OpenAI, citing a recent interview with Nadella. Amid a dramatic leadership shake-up at OpenAI late last year, Nadella stated that if OpenAI disappeared tomorrow we have all the IP rights and all the capability. We have the people, we have the compute, we have the data, we have everything. We are below them, above them, around them. The lawsuit presents this as evidence that OpenAI has strongly served Microsofts interests.

The lawsuit also centers around OpenAIs GPT-4, which Musk claims constitutes AGI an AI whose intelligence is at par, if not higher, than humans. He alleges OpenAI and Microsoft have improperly licensed GPT-4 despite agreeing that OpenAIs AGI capabilities would remain dedicated to humanity.

Through the lawsuit, Musk is seeking to compel OpenAI to adhere to its original mission and bar them from monetizing technologies developed under its nonprofit for the benefit of OpenAI executives or partners like Microsoft.

The suit also requests the court rule that AI systems like GPT-4 and other advanced models in development constitute artificial general intelligence that reaches beyond licensing agreements. In addition to injunctions forcing OpenAIs hand, Musk asks for accounting and potential restitution of donations meant to fund its public-minded research should the court find it now operates for private gain.

Mr. Altman hand-picked a new Board that lacks similar technical expertise or any substantial background in AI governance, which the previous board had by design. Mr. DAngelo, a tech CEO and entrepreneur, was the only member of the previous board to remain after Mr. Altmans return. The new Board consisted of members with more experience in profit-centric enterprises or politics than in AI ethics and governance, the lawsuit adds.

Read this article:

Elon Musk sues OpenAI and Sam Altman over betrayal of nonprofit AI mission - TechCrunch

Google launched a private program for a handful of independent publishers last month, providing the news organizations with beta access to an unreleased generative artificial intelligence platform in exchange for receiving analytics and feedback, according to documents seen by ADWEEK.

As part of the agreement, the publishers are expected to use the suite of tools to produce a fixed volume of content for 12 months. In return, the news outlets receive a monthly stipend amounting to a five-figure sum annually, as well as the means to produce content relevant to their readership at no cost.

In partnership with news publishers, especially smaller publishers, were in the early stages of exploring ideas to potentially provide AI-enabled tools to help journalists with their work,

This speculation about this tool being used to re-publish other outlets work is inaccurate, a Google representative said in a statement. The experimental tool isbeing responsiblydesigned to help small, local publishers produce high quality journalism using factual content from public data sourceslike a local governments public information office or health authority. These tools are not intended to, and cannot, replace the essential role journalists have in reporting, creating and fact-checking their articles.

The beta tools let under-resourced publishers create aggregated content more efficiently by indexing recently published reports generated by other organizations, like government agencies and neighboring news outlets, and then summarizing and publishing them as a new article.

Other gen AI experiments Google has released over the past two years include the codenamed Genesis, which can reportedly produce whole news articles and was privately demonstrated to several publishers last summer, according to The New York Times. Others, including Search Generative Experience and Gemini, are available for public use and threaten to upend many of the commercial foundations of digital publishing.

The program is part of the Google News Initiative, which launched in 2018 to provide publishers with technology and training.

Although many of its programs indisputably benefit the publishers involved, the broader reception of GNI has been mixed.

Google has used GNI to drum up positive press and industry goodwill during moments of reputational duress, and many of the commercial problems it aims to solve for publishers were created by Google in the first place, said Digital Content Next CEO Jason Kint.

The larger point here is that Google is in legislative activity and antitrust enforcement globally for extracting revenue from the publishing world, Kint said. Instead of giving up some of that revenue, its attacking the cost side for its long-tail members with the least bargaining power.

Google first shared a call for news organizations to apply to test the emerging technologies in an October edition of the Local Independent Online News newsletter.

GNI began onboarding publishers in January, and the yearlong program kicked off in February.

According to the conditions of the agreement, participating publishers must use the platform to produce and publish three articles per day, one newsletter per week and one marketing campaign per month.

To produce articles, publishers first compile a list of external websites that regularly produce news and reports relevant to their readership. These sources of original material are not asked for their consent to have their content scraped or notified of their participation in the processa potentially troubling precedent, said Kint.

When any of these indexed websites produce a new article, it appears on the platform dashboard. The publisher can then apply the gen AI tool to summarize the article, altering the language and style of the report to read like a news story.

Its hard to argue that stealing peoples work supports the mission of the news.

Digital Content Next CEO Jason Kint

The resulting copy is underlined in different colors to indicate its potential accuracy: yellow, with language taken almost verbatim from the source material, is the most accurate, followed by blue and then red, with text that is least based on the original report.

A human editor then scans the copy for accuracy before publishing three such stories per day. The program does not require that these AI-assisted articles be labeled.

The platform cannot gather facts or information that have not already been produced elsewhere, limiting its utility for premium publishers.

Articles produced by the platform could also draw traffic away from the original sources, negatively affecting their businesses. The process resembles the ripping technique newly in use at Reach plc, except that in this case, the text is pulled from external sources.

I think this calls into question the mission of GNI, Kint said. Its hard to argue that stealing peoples work supports the mission of the news. This is not adding any new information to the mix.

Go here to read the rest:

Google Is Paying Publishers Five-Figure Sums to Test an Unreleased Gen AI Platform - Adweek

Researchers using an AI photo-scanning tool similar to facial recognition have learned that theres been a 20% decline in North Pacific Ocean humpback whale populations over the past decade. The researchers pointed to a climate change related heat wave as a possible culprit. The findings, published this week in Royal Society Open Science, used the artificial intelligence-powered image detection model to analyze more than 200,000 photographs of humpback whales taken between 2001 and 2022.

Facial recognition models used to identify humans have faced sustained criticism from researchers and advocates who say the models struggle to identify accurately identity nonwhite people. In this case, the model scanning humpback whale photos was trained to spot and recognize unique identifiers on a whales dorsal fin. These identifiers function like a one-of-a-kind whale fingerprint and can consist of marks, variations in pigmentation, scarring, and overall size. Researchers used successful photo matches to inform estimates for humpback whale populations over time.

[ Related: The government is going to use facial recognition more. Thats bad. ]

Images of the whale tails, captured by scientists and whale watchers alike, are stored by a nonprofit called HappyWhale, which described itself as largest individual identification resource ever built for marine mammals. HappyWhales encourages everyday citizen scientists to take photos of whales they see and upload them to its growing database. The photos include the data, and location of where the whale was spotted.

From there, users can track a whale they photographed and contribute to a growing corpus of data researchers can use to more accurately understate the species population and migration patterns. Prior to this AI-assisted method, experts had to comb through individual whale tail photographs looking for similarities with their named eye, a process both painstaking and time-consuming. Image matching technology speeds that process, giving researchers more time to investigate changes in population data.

Having an algorithm like this dramatically speeds up the information-gathering process, which hopefully speeds up timely management actions, Philip Patton, a University of Hawaii at Manoa Phd student who has worked with the tool said in a previous interview with Spectrum News.

Humpback whales, once on the brink of extinction, have seen their population grow in the 40 years since commercial hunting of the species was made illegal, so much so that the giant mammals were removed from the endangered species list in the US in 2016. But that rebound is at risk of being short-lived. Researchers analyzing the whale data estimate their population peaked in 2012 at around 33,488. Then, the numbers started trickling downwards. From 2012 to 2021, the whale population dropped down to 26,662, a decline of around 20%. Researchers say that downward trend coincided with a record heat wave that raised ocean temperatures and may have altered the course of species recovery.

That historic heat wave resulted in rising surface sea temperatures and decreases in nutrient-rich water which in turn led to reductions in phytoplankton biomass. These changes led to greater disruptions in the food chain which the researcher says limited the whales access to krill and other food sources. While they acknowledged ship collisions and entanglements could be responsible for some of the population declines, the researchers said those factors couldnt account for the entirety of the decline.

These advances have shifted the abundance estimation paradigm from data scarcity and periodic study to continuous and accessible tracking of the ocean-basin- wide population through time, the researchers wrote.

Whales arent the only animals having their photos run through image detection algorithms. Scientists use various forms of the technology to research populations of cows, chickens, salmon, and lemurs, amongst species. Though primarily used as an aid for conservation and population estimation, some researchers have reportedly used the technology to analyze facial cues in domesticated Sheep to determine whether or not they felt pain in certain scenarios. Others have used photo matching software to try and to find missing pets.

[ Related: Do all geese look the same to you? Not to this facial recognition software. ]

These examples and others highlight the upside of image and pattern matching algorithms capable of sifting through vast image databases. In the case of conservation, accurate population estimates made possible by these technologies can help inform whether or not certain species require endangered classifications or other resources to help maintain their healthy population.

See the original post here:

Why scientists are tracking whale tails with AI - Popular Science