OpenAI has unveiled its latest artificial intelligence system, a program called Sora that can transform text descriptions into photorealistic videos. The video generation model is spurring excitement about advancing AI technology, along with growing concerns over how artificial deepfake videos worsen misinformation and disinformation during a pivotal election year worldwide.

The Sora AI model can currently create videos up to 60 seconds long using either text instructions alone or text combined with an image. One demonstration video starts with a text prompt that describes how 鈥渁 stylish woman walks down a Tokyo street filled with warm glowing neon and animated city signage鈥�. Other examples include a dog frolicking in the snow, vehicles driving along roads and more fantastical scenarios such as sharks swimming in midair between city skyscrapers.

鈥淎s with other techniques in generative AI, there is no reason to believe that text-to-video will not continue to rapidly improve 鈥� moving us closer and closer to a time when it will be difficult to distinguish the fake from the real,鈥� says at the University of California, Berkeley. 鈥淭his technology, if combined with AI-powered voice cloning, could open up an entirely new front when it comes to creating deepfakes of people saying and doing things they never did.鈥�

Sora is based in part on OpenAI鈥檚 preexisting technologies, such as the image generator DALL-E and the GPT large language models. Text-to-video AI models have lagged somewhat behind those other technologies in terms of realism and accessibility, but the Sora demonstration is an 鈥渙rder of magnitude more believable and less cartoonish鈥� than what has come before, says , co-founder of SocialProof Security, a white-hat hacking organisation focused on social engineering.

To achieve this higher level of realism, Sora combines two different AI approaches. The first is a diffusion model similar to those used in AI image generators such as DALL-E. These models learn to gradually convert randomised image pixels into a coherent image. The second AI technique is called 鈥渢ransformer architecture鈥� and is used to contextualise and piece together sequential data. For example, large language models use transformer architecture to assemble words into generally comprehensible sentences. In this case, OpenAI broke down video clips into visual 鈥渟pacetime patches鈥� that Sora鈥檚 transformer architecture could process.

Sora鈥檚 videos still contain plenty of mistakes, such as a walking human鈥檚 left and right legs swapping places, a chair randomly floating in midair or a bitten cookie magically having no bite mark. Still, , a senior research scientist at NVIDIA, took to the social media platform X to praise Sora as a 鈥渄ata-driven physics engine鈥� that can simulate worlds.

The fact that Sora鈥檚 videos still display some strange glitches when depicting complex scenes with lots of movement suggests that such deepfake videos will be detectable for now, says at Princeton University. But he also cautioned that in the long run 鈥渨e will need to find other ways to adapt as a society鈥�.

OpenAI has held off on making Sora publicly available while it performs 鈥渞ed team鈥� exercises where experts try to break the AI model鈥檚 safeguards in order to assess its potential for misuse. The select group of people currently testing Sora are 鈥渄omain experts in areas like misinformation, hateful content and bias鈥�, says an OpenAI spokesperson.

This testing is vital because artificial videos could let bad actors generate false footage in order to, for instance, harass someone or sway a political election. Misinformation and disinformation fuelled by AI-generated deepfakes ranks as a major concern in academia, business, government and other sectors, as well as .

鈥淪ora is absolutely capable of creating videos that could trick everyday folks,鈥� says Tobac. 鈥淰ideo does not need to be perfect to be believable as many people still don鈥檛 realise that video can be manipulated as easily as pictures.鈥�

AI companies will need to collaborate with social media networks and governments to handle the scale of misinformation and disinformation likely to occur once Sora becomes open to the public, says Tobac. Defences could include implementing unique identifiers, or 鈥渨atermarks鈥�, for AI-generated content.

When asked if OpenAI has any plans to make Sora more widely available in 2024, the OpenAI spokesperson described the company as 鈥渢aking several important safety steps ahead of making Sora available in OpenAI鈥檚 products鈥�. For instance, the company already uses automated processes aimed at preventing its commercial AI models from generating depictions of extreme violence, sexual content, hateful imagery and real politicians or celebrities. With more people than ever before , those safety steps will be crucial.

THE enthusiastic Simon Clark won鈥檛 stop reeling off a list of his favourite science and technology YouTubers. There is Marques Brownlee, a reviewer who gets his hands on consumer tech before most mainstream magazines and websites. There is Flammable Maths, a quirky mathematician whose videos are the comedic utterings of a troubled genius. And there is Tim Blais, under the moniker A Capella Science, who explains evolutionary development to the tune of the song Despacito, a hit for Justin Bieber.

Sometimes you don鈥檛 know what you don鈥檛 know until you find it. And Clark鈥檚 list of fellow video-makers is a good place to hunt for serendipitous insights on YouTube. The video-sharing website is now the second-most visited platform on the internet, ahead of Facebook and behind Google. The site claims to host more educational videos than there are books in the US Library of Congress, and that more than a billion hours of videos are watched every day. And within that giant pool, thousands of creators-cum-scientists are putting out educational content seen by millions.

Among them is In茅s Dawson, who recently completed her PhD on the biomechanics of insect flight. She is a respected science communicator who has been posting videos on YouTube for two years.

Dawson set up her YouTube channel during a slump in her PhD. 鈥淓veryone seems to go through it,鈥� she says. 鈥淭hings aren鈥檛 really working out and people start to get a bit down about how their work is going.鈥� She hoped dabbling in science communication would rekindle her passion. It did. Now she is shunning traditional academia, combining her YouTube channels (one in Spanish, one in English) with a freelance career in science communication.

鈥淚 feel like I鈥檓 sharing the joy in wondering about something and finding something out,鈥� she says. Each of the videos takes her a day to script, a few hours to shoot, then half a day to edit, with her trademark animations taking about 2 hours to produce 30 seconds of footage. They aren鈥檛 didactic, but she hopes the videos will prompt viewers to consider their questions about the world, then go down the YouTube rabbit hole to look for answers. 鈥淭here鈥檚 such a plethora of content that there鈥檚 literally something for everyone,鈥� she says.

鈥淢ore than 1.5 million people watch Simone Giertz build odd robots and equipment鈥�

Even the illicit. Drugslab, which aims to educate about drug use rather than to encourage it, investigates the chemical composition and biological effects of drugs. The three hosts take narcotics in controlled conditions, and delve into the science of what is going on as they trigger hallucinations, quicken pulses and dilate pupils.

Or perhaps you would rather channel your inner inventor? Simone Giertz is your best bet. She dropped out of a physics course and is now the self-styled 鈥渜ueen of shitty robots鈥�, and describes her channel as 鈥渓ess educational than it is inspirational鈥�. More than 1.5 million people watch her build odd robots and equipment.

Giertz considered making her content more educational, but says: 鈥淭here are so many people already doing that really well who have much more authority than I do.鈥� Instead, she is broadening out her content to make it a 鈥渏ournal of personal interest鈥�, looking at how things work 鈥� from electric vehicles to orchids.

As well as having his own favourite YouTubers, Clark 鈥� who got a physics degree at the University of Oxford 鈥� is also followed by almost 200,000 viewers. He started his channel in 2010, providing advice for those looking to apply to Oxford. As a state school alumnus, that was something Clark couldn鈥檛 find when he needed it. He began creating science videos: one video about which planets in the Star Wars universe could and couldn鈥檛 exist proved particularly popular. (Yes, Tatooine can exist.)

Like Dawson, Clark pitches his videos in a particular way. 鈥淭hey鈥檙e halfway between a video I鈥檇 want to watch and a paper,鈥� he says. Some 鈥渆ducational鈥� content on YouTube is more like a mere collection of facts. Clark, Dawson and many others piece together information and draw conclusions, showing their scientific method along the way.

鈥淚 don鈥檛 normally make videos on topics I鈥檝e studied in my PhD, but I do download and find several papers relevant to the topic,鈥� says Dawson. 鈥淚n some ways, it鈥檚 equivalent to my undergraduate degree 鈥� you get a reading list, then have to argue to justify your thoughts on the topic.鈥� Clark includes citations to academic papers on screen 鈥� a flagging up method adopted by other educational YouTubers since.

But science on YouTube suffers from a gender problem. A recent analysis of nearly 400 of the most popular science, engineering and maths channels on the platform showed that fewer than 10 per cent were hosted by women.

That is even worse than in the wider scientific community, where just 30 per cent are women. And those channels that were hosted by women were subjected to far more hostile, sexist and critical comments under the videos.

The gender gap is also evident in who is watching. 鈥淚t鈥檚 all men on my channel,鈥� says Giertz, and mostly aged 25 to 34. 鈥淎ll my comment sections are engineers at Google.鈥�

Most of Dawson鈥檚 viewers are also aged 25 to 34, with a significant increase in those in secondary/high school and university watching lately. Yet more than 80 per cent of her viewers are men. 鈥淚 have a theory that a lot of this is algorithmically determined,鈥� she says. This hunch is supported by the fact that YouTube recommends make-up tutorials to her, even though she never watches them, and these are served up more regularly than scientific ones.

That said, things are changing 鈥� slowly, with a more equal split among her younger viewers. Hopefully, the 2030 version of YouTube 鈥� and the academic community 鈥� will look very different.