Welcome to our weekly roundup of premium handcrafted AI news.
This week AI voice clones got banned but became super sneaky.
Advanced AI added to anxious artists’ angst.
And Indonesian politicians brought a dead dictator back to life.
Let’s dig in.
Misaligned
The fight to make sure AI behaves better is ongoing. The FCC declared that AI-generated voices in robocalls are illegal, so ‘Joe Biden’ probably won’t be calling you again.
Voice cloning is getting so good it’s hard to tell who we’re speaking to anymore. IBM Security made a very sneaky proof of concept to show how AI can hijack a key part of the audio in a conversation. This is pretty scary.
Does AI display racial and gender bias when evaluating images? Researchers found that it sometimes does, but they wonder if it’s always a bad thing.
If we did eventually make a perfectly aligned AI that was completely ‘safe’, what would that be like? Meet Goody-2, the world’s most responsible AI model. I’m offended.
Get creative
Can AI be creative? Is AI-generated art really art? Sam takes a deep dive into how artists are under fire and investigates the impact of AI on creativity and jobs in the sector.
This Sunday, more than a third of the US population watched the Super Bowl. A lot of those viewers tuned in purely for the commercials. AI dominated the Super Bowl ads and gave us some insight into how folks feel about the tech.
McAfee released its second “Modern Love” study, investigating the interaction between AI and romance. Yes, that’s a thing now. Users of online dating platforms are getting creative and using AI to juice their profiles and scam people.
Is the size of your dataset making you feel a little inadequate? MIT researchers published a fascinating study that shows that symmetry of data could solve small dataset woes.
Chasing chips
There doesn’t seem to be any end to the insatiable demand for high-end AI chips. OpenAI CEO Sam Altman wants to make his own and says he needs trillions (with a t) for his outlandish AI chip project.
While Altman dreams of his own AI chip factories, NVIDIA CEO Jensen Huang is advising countries to build their own sovereign AI assets. And he knows just where they could buy the GPUs they’ll need.
We may not need to pay for cloud computing or API calls soon. NVIDIA has released Chat with RTX, a demo that enables you to run an LLM like Llama 2 or Mistral locally on your Windows PC using one of its RTX GPUs.
AI across the globe
Microsoft and OpenAI discovered that a number of high-profile cybercrime groups were using their AI tools and promptly blocked their accounts.
They say these groups are backed by nation-states like North Korea, Iran, Russia, and China. Even the names of some of these cybercrime groups sound a little phishy. One group is called Salmon Typhoon. I wish I was making this up.
I had ChatGPT make a pic they could use on their t-shirts for their next cybercrime office team-building day. Just kidding guys. Please don’t hack me.
Geopolitical tensions are influencing where AI investment flows and who you can do business with. Abu Dhabi-based AI firm G42 has cut ties with Chinese firms after coming under pressure from the US government and investors.
If you’re trying to make sense of how to use AI to market your business globally, then sign up for the upcoming MarTech Summit events that will cover the APAC and EMEA regions.
Meanwhile, China denies any nefarious intentions with its pursuit of AI tech. A few days ago Chinese researchers unveiled a robot toddler named “Tong Tong” which they say is as intelligent as a child aged between three and four.
China says it wants to mass-produce humanoid robots by 2025. Millions of robotic toddlers, what could possibly go wrong?
The King’s Cross
Fans of British royalty got a dose of AI fakery this week. AI-generated books are flooding Amazon’s shelves to speculate about King Charles’s cancer diagnosis. The folks at Buckingham Palace are not amused.
AI surveillance technology was used in a year-long trial at a London Underground station with some promising results. The tech is likely to be rolled out wider but the line between safety and Big Brother surveillance is a fine one.
Jailbird and Zombie politicians
Politics is always entertaining but this week it got extra weird.
Ex-Pakistan Prime Minister Imran Khan is in prison but his party was still contesting the recent elections. He just declared that he won the elections using an AI avatar and a clone of his voice.
The US may have some of the oldest presidential candidates but, not to be outdone, an Indonesian party used AI to resurrect a dead dictator for election propaganda. Former President Suharto looked pretty good in this video, considering he’s been dead for 16 years.
Some of the best cartoons and posters in history were inspired by political satire or propaganda. But because of the risk of hyper-realistic AI fakes, Midjourney is considering snubbing out AI-generated images of Trump or Biden as the election looms.
The AI doctor will see you now
You could spend a few hundred dollars an hour on a therapist, or you could just buy an Apple Vision Pro. Researchers developed an immersive AI therapy system named XAIA which combines GPT-4 with Apple’s new VR headset.
Participants in the trial gave positive feedback on the experience. It could also be that they were just happy to get to play with an Apple Vision Pro without the $3,500 price tag.
Apple Vision Pro pic.twitter.com/06KO0b74Er
— Aleksandr Morozov (@almormd) June 5, 2023
Finding the right patients to participate in a clinical trial involves wading through thousands of pages of clinical notes and records. These researchers found a way to use GPT-4V to do this faster, better, and a lot cheaper than humans can.
In other news…
Here are some other clickworthy AI stories we enjoyed this week:
- The White House wants to ‘cryptographically verify’ videos of Joe Biden so viewers don’t mistake them for deepfakes.
- Apple buys iWork.ai domain as evidence of a big AI push mounts.
- $1.72 trillion AI chip giant NVIDIA is about to pass Amazon as the 4th-most valuable U.S. company.
- After all the hype, Gemini Ultra is now available as part of a rebranded Gemini subscription option.
- Innovation is outpacing our ability to handle misinformation on social media as elections loom.
- The judge in the OpenAI copyright case dismissed the bulk of the authors’ copyright claims.
And that’s a wrap.
After seeing IBM Security’s voice clone hack I think we’ll all need to become super skeptical real fast.
How do you feel about bringing back dead presidents to help Biden and Trump campaign? Who’d you like to see in action? I’m thinking, Andrew Johnson after a few drinks debating Richard Nixon. I’d watch that.
Altman asking for trillions and everyone nodding in agreement instead of calling the nuthouse to make a reservation for him sums up how crazy the AI scene has become over the last year.
We can’t wait to see what happens next. And when it does, we’ll let you know right here.