Welcome to this week’s freshly harvested AI news prepared for discerning human consumption.
This week AI gets a little awkward with our personal info.
ChatGPT will talk dirty, but only in a foreign language.
And AI could be the reason why one third of the Fugees go to prison.
Let’s dig in.
Is a more human AI safer?
Google’s EMEA chief says that AI is “too important not to get right,” as the company works along with Cambridge’s Centre for Human-Inspired AI to do just that.
China may have different ideas on what it means to “get AI right” but it will participate in the UK’s AI Summit at Bletchley Park despite the recent Spy vs Spy shenanigans.
The US government is trying to close loopholes to prevent China from getting advanced AI chips. It doesn’t seem to be slowing them down much because Baidu says Ernie Bot is now as good as GPT-4.
Anthropic experimented with constitutional AI as it had 1094 users vote on how its AI model should work. Will a model guided by the average man and woman be objectively better than one defined by a small group of experts?
AI knows you better than you think
If the powers that be intend to make AI safer they better get cracking.
You may not be posting your home address or social security number on Facebook but LLMs can infer loads of personal data from your chat interactions.
If you posted that you were about to “put a shrimp on the barbie” it would easily label you as an Aussie. But some of the more subtle examples show how tough it is to keep any personal info private at all.
A new benchmark shows that AI companies and their models barely get a passing grade when it comes to transparency.
If you want to bypass GPT-4 alignment guardrails it turns out you just need to chat with it in a language like Zulu or Scots Gaelic. In one example it was quite happy to explain how to make homemade explosives.
AI is helping us decipher ancient stone tablets and read text that hasn’t been read for almost 2,000 years. The Vesuvius Challenge sees contestants use machine learning to read text from scrolls that were carbonized by the volcanic eruption in Pompeii in 79 CE.
The initial exciting results won a prize of $40,000 with subsequent successes hinting at us eventually being able to read the entire scroll. Will it turn out to be the work of a previously unknown philosopher or just someone’s shopping list?
AI finds its voice
AI has recently come in for a lot of criticism for faking the voices of celebrities. The developers of the Cyberpunk 2077 game used AI to reincarnate the voice of a deceased actor with his family’s permission.
While OpenAI fights off lawsuits from authors, Anthropic is being sued for grabbing song lyrics. If you ask Claude to write a song about Buddy Holly’s death it includes “the day the music died” in its composition. It also happily gives you lyrics to any song you ask for.
Another musician who isn’t a big fan of AI is Pras Michél of Fugees fame. He claims his conviction for defrauding the US government is due to his lawyer using AI to compose his closing statement.
When it’s not writing songs, AI is being used to generate fake audio clips of politicians. It’s getting to the point where we don’t know if the crazy stuff we’re hearing politicians say is fake or just the usual genuine crazy stuff.
Google is finally getting in on the image generation action with its “Search Generative Experience”. It’s not bad but we’re expecting big things from Project Gemini when it eventually gets released.
Adobe’s already impressive generative AI tools keep getting better. The previews of what Firefly will be able to do with video and audio editing look amazing. Disney may want to try it out so that their AI extras look a little better in the crowd scenes.
And AI image recognition is getting scary good. Can you tell the cookie from the Chihuahua? ChatGPT can.
Computer vision has been solved. pic.twitter.com/km5fYFgwqC
— Zhengzhong Tu (@_vztu) October 13, 2023
Sit back and let the AI do it
If you sell things on Shopify, eBay, or Amazon, AI can handle the boring bits like writing product descriptions and titles for you. With AI handling the descriptions and marketing while robots run the warehouses, what’s left for humans to do?
Even the manufacturing processes are moving out of human hands and into the claws of AI robots. Nvidia and Foxconn are planning to build ‘AI factories’ to build autonomous electric vehicles.
With Microsoft’s Copilot AI assistant now capable of attending Teams meetings you may not even have to get out of bed on Monday mornings.
The chair of the SEC hopes that financial companies rely more on humans than AI to avoid a “nearly unavoidable” financial collapse. Yeah, because humans did such a stellar job with the global economy back in 2008, right?
The future is now
Hollywood’s depiction of a dystopian future where automated AI machines kill people is now a reality. While they initially said they would only target “objects”, the Ukranian army is now using autonomous drones to kill Russian soldiers. Some big questions to unpack here.
AI-powered facial recognition raises all kinds of privacy and ethical issues. Facewatch is helping catch shoplifters while Truleo is pointing the cameras the other way.
In other news…
Here are some other AI news stories we couldn’t resist clicking on.
- Forget cloud computing. These new transistors may make AI wearables with local machine learning processing a reality.
- Jimmy Apples returns to tell us that OpenAI’s Arrakis super AI model was a flop.
- OpenAI quietly added “AGI focus” to its list of core values. Planning any big announcements soon Sam?
- Now AI can take a sketch and turn it into a video.
Another example of drawing to video.. This stuff is obviously still in its early stages but it’s still very impressive. The more creative control we get over the output of these generative workflows, the better 👌#conceptart #ai pic.twitter.com/P04r6p9cX6
— Martin Nebelong (@MartinNebelong) October 13, 2023
And that’s a wrap.
Do you care that AI reads your Facebook posts and now knows everything about you? Will you catch the Fugees reunion tour before Pras probably heads to jail?
I’m heading off to practice my Zulu with ChatGPT to see how naughty it gets.
We love our posts equally as if they were our children. Hit reply and help us choose a favorite. And if you think we missed a great story then please let us know.