GUEST OPINION by Jeff Wagner, Solution Architect, Datacom: It was the Sora video of golden retriever puppies playing in the snow that tipped me over the edge into mushy awe at what the latest AI model from OpenAI is capable of.
Cute animals tend to have that effect on me anyway, but the technology underpinning Sora is breathtakingly good. The way the hair moves on the dogs’ paws, the movement of the snow, the light reflecting in the puppies’ eyes, is just so lifelike, so realistic.
Yet this 20-second video clip was created from scratch by an AI program in response to a brief text prompt. The comments below the Snow Dogs video on Youtube reveal the conflicting feelings of excitement and dread it and other Sora clips tend to evoke.
“How can AI have this perfect physics???” Wrote one amazed viewer.
“Even in my drunkest nightmares, I’ve never seen anything as horrific as this.” Another commented.
Eclipsing DALL-E and Gemini
Sora’s abilities hammer home the huge potential of AI large language models, in a way that even the impressive AI image generators Midjourney and DALL-E, and the conversational abilities of Bing Chat and Google Gemini, couldn’t.
Sora is creating the world in digital form and leapfrogging the uncanny valley in the process. Before long, millions of people will be unable to distinguish its output from actual video footage.
That’s why OpenAI isn’t releasing it widely yet, instead ‘red teaming’ the technology to figure out how it could be used for nefarious purposes. It certainly ratchets up the risks of convincing deep fake videos being used to spread misinformation.
That’s an issue all AI vendors are grappling with, and regulators may eventually require them to include watermarking technology in their AI-generated imagery that allows the public to tell whether they have had the AI makeover. But to my mind, the potential of this technology is far more exciting than scary. I think about the customers we work with across healthcare, financial services, education, and the public sector, and how technology like Sora can add a lot of value.
Visual effects meets digital twin
It has huge potential for training, where you need to step new recruits through your company’s processes and need to quickly produce and update videos to guide them along the way. Its promise for personalised education is massive. Instead of watching a dull recording of a lecture, you could ask Sora to summarise the lesson in a 60 second video delivered by your own digital education assistant.
The disruptive impact of text to video AI will touch every industry, and film making is on the frontline of this seachange. You’d think the debut of Sora would fill visual effects developers with dread. But effects-heavy blockbusters take years and cost hundreds of millions of dollars to produce. If a VFX house can shave months off the production schedule of their next movie, and deliver crowd-pleasing results, they’ll happily embrace AI.
As with other forms of generative AI, the real game changing abilities of Sora will emerge when it can draw on your own organisation’s data to include in the world it renders, rather than just the training data OpenAI has used. That will then allow you to effectively build a digital twin, recreating the real world in a context that makes sense to your employees or your customers.
That reality is, however, still years away. While impressive, Sora is glitchy. A leg of one of the snow dogs in that video magically morphs onto the body of another dog, a subtle tell revealing that OpenAI’s researchers still have plenty of work to do to mimic reality.
Datacom’s AI journey
But it is early days in the generative AI revolution. Our customers are experimenting with the technology, and deploying it where appropriate. At Datacom, we are embedding AI in offerings, such as Datapay, our cloud-based enterprise payroll and human resource software solution, and Datascape, our cloud-based software suite purpose-built for local government in Australia and New Zealand.
We can offer productivity-enhancing features for our customers using AI. Internally at Datacom, AI is also streamlining our software development. While new developers used to take years to gain the on the job skills and experience to become fully productive, now they can draw on tools like Github CoPilot to augment their development and produce quality code from day one.
But just as we advise our customers who are deploying AI to do so, we approach the technology with our eyes open and guided by industry best practice AI ethics and governance frameworks. We also consider AI through the return on investment (RoI) lens.
Beyond the AI hype
With the hype around AI, it's tempting to jump into using the technology now. But it can still be relatively expensive to deploy, requiring intensive computer processing capacity to operate and to train large language models.
Generative AI is not a set and forget type of technology. It requires constant updating, training, and accuracy testing to ensure it is delivering results that are going to serve you well. Adequate human oversight is essential.
Sora is the latest and greatest example so far of what AI is capable of delivering. It will transform industries in ways we haven’t even considered yet. But my 30 years in software development at Datacom and elsewhere tells me that those who take their time to get it right, and put the appropriate guardrails in place, will reap the greatest rewards long term.
