It's Been a Rollercoaster of a Week in AI
From a ChatGPT meltdown and eye-popping AI-generated videos, to chip-manufacturer NVIDIA's market value going through the roof, AI's hardly been out of the spotlight this past few days
I don’t usually do roundup articles, but there was so much stuff happening in AI this past few days that I couldn’t resist.
As a result, here are five things in the ever-changing world of AI that grabbed my attention this past week — with commentary of course!
Meet Sora — OpenAI’s AI Video Generator
If you’ve been anywhere near social media over the past few days you can’t have failed to miss the hype around OpenAI’s new platform for AI-generated video: Sora.
Sora was announced on Feb 15 (although it isn’t available for general use yet) and has been creating rather substantial waves.
The videos attributed to Sora are stunning (it’s worth checking out the ones here as well as on the video above). Of course they suffer from reality glitches, much as early AI image generators had a tendency to give people too many fingers. But at first glance they’re incredibly realistic or, in the case of the more fantastical ones, amazingly captivating.
As many testers have shown, what makes this platform stand out is that users can very simply use it to create engaging videos of scenes that range from emulating real life to the utterly fantastical — all with a few lines of text.
It’ll be interesting to see what people do with Sora when it becomes more widely available. But this could be another Generative AI tool that unleashes the imagination and capability of people of all ages and abilities who struggle to translate what they can imagine in their heads into visual media to share with others.
Of course, it also raises the usual questions around bias, intellectual property and source material — as well as how this might impact jobs. The biggie though is the growing capacity for nearly anyone to create misleading deep fakes (or deepfakes as they are more commonly known).
OpenAI are watermarking videos to try and avoid this — but how many people will be looking for that lower-right watermark that warns them what they are looking at is not real?
Disrupting the Deepfake Supply Chain
Talking of deepfakes, an open letter was published this past week calling for new laws to fully criminalize deepfake child pornography, establish criminal penalties for anyone who knowingly creates or facilitates the spread of harmful deepfakes, and to require software developers and distributors to prevent their audio and visual products from creating harmful deepfakes — and to be held liable if their preventive measures are too easily circumvented.
As of typing, over 1000 people have already signed the letter, including many prominent names in AI (including Andrew Yang, Marietje Schaake, Stephen Pinker, Stuart Russel, Gary Marcus, and Gillian Hadfield).
I tend not to sign open letters like this as they’re often deeply naive when it comes to how new technologies are governed. But in this case I did.
Several years ago I was asked at an event how serious I thought deepfakes were. At the time I was of the opinion that most people would use their common sense to differentiate between what was fake and what was not.
Today I am far less sure — especially when it comes to uses that promote deeply unethical predilections such as the generation of child porn, and that have the capacity to destabilize relationships, communities, and whole societies.
And so I added my signature to the list.
If you’re interested in adding yours, the petition can be accessed here.
Google Finds Itself in AI Deep Water … Again
Google has a venerable history of generative AI launch fails (remember the launch of Bard?) And this past week was no exception.
A couple of weeks ago, Google renamed Bard Gemini. So far so good. Then this past week stories started circulating about Gemini being so eager to show racial diversity in generated images that historical figures were being replaced by people who were most definitely not true to type. (Think “racially diverse Nazis” from WWII)
As outrage spread on social media (including Elon Musk weighing in), Google apologized and paused the GenAI platform’s ability to generate images.
I feel some sympathy here for Google. For sure the fine tuning of Gemini went off the rails a little. But Google are clearly trying to create a product that is socially responsible, and were fast to address concerns.
My sense is that this was more of a “gotcha” moment for many people than a dangerous flaw. But the incident did show how hard it is to constrain such complex generative models, and how easy it is for users to poke holes in them.
And truth be told, it’s impossible to get a generative AI system 100% perfect before it launches — meaning that it’s far more important to be agile and responsive when things do go awry.
And good on Google for doing just this.
ChatGPT Throws a Wobbly
Google wasn’t the only company feeling the AI heat this past week. On February 21 reports started to appear on social media that OpenAI’s ChatGPT was having a bit of a wobbly.
OpenAI acknowledged the issue and rapidly corrected it.
Of course this was just a one-off event, and I’m sure that ChatGPT is now feeling much better and behaving normally. However, the incident did serve as a reminder that useful as generative AI platforms like ChatGPT are, they are neither robust, reliable, or trustworthy.
On the other hand, maybe the ability to have an off day is in itself another indicator of emergent of human like behavior …
And finally, NVIDIA comes a step closer to total AI Hardware Dominance
The big AI hardware news this past week was that chip and processor manufacturer NVIDIA topped an impressive $2 Trillion market value — showing a meteoric rise from a somewhat niche processor manufacturer to one of the most important AI chip manufacturers in the world.
To put this into context, Amazon currently stands at $1.82 Trillion, Alphabet (Google’s parent) at $1.80 Trillion, and Intel at a mere $182 billion.
NVIDIA started life in the 1990’s designing and producing 3D graphics technology. In 1999 they invented the Graphics Processing Unit (GPU) to power high-level graphical processing — and became the darling of high end gamers.
Then it was discovered that GPUs are ideal for high performance computing, and especially AI, and everything changed.
NVIDIA’a rise to dominance in the AI hardware arena — and the sheer size of the company’s current market value — is a sobering reminder of just how fast AI is accelerating. Unlike software, hardware takes much longer to develop and perfect. And so investments here are a persuasive indicator of how AI is likely to continue to grow in scale and impact in the near future.
Whether NVIDIA can retain its AI chip pole position is unclear — especially after Sam Altman’s audacious plans to raise $7 Trillion for AI chip development. But in the meantime, check out this LinkedIn post to see just how fast NVIDIA’s AI market share has grown (wait to the end to see the shocker).