- Loop
- Posts
- 🇺🇳 UN creates a new AI safety group, while OpenAI create a doomsday team
🇺🇳 UN creates a new AI safety group, while OpenAI create a doomsday team
Plus more on Amazon’s GenAI tools for advertisers, Cruise’s robotaxi struggles, and computer vision company Viso.
Liam
October 30, 2023
Hello,
Welcome to this edition of Loop! We aim to keep you informed about technology advances, without making you feel overwhelmed.
To kick off your week, we’ve rounded-up the most important technology and AI updates that you should know about.
In this edition, we’ll explore:
- UN’s new AI regulation group
- OpenAI’s doomsday team
- GenAI tools from Amazon & Google
- … and much more
Let's jump in!
Top Stories
1. OpenAI creates AI doomsday team [Link]
The new “Preparedness” team aims to predict and mitigate dangers from future AI systems. They will cover areas from deception and malicious code generation - to much more serious threats, such as using AI to make chemical or nuclear weapons. The team will develop a "risk-informed development policy" to guide AI model evaluations, monitoring, and governance.
Of course, this announcement coincides with the UK Government’s AI safety summit - which is running this week and involves most of the top AI companies.
2. Google is working on a GenAI tool that quickly builds app prototypes [Link]
Leaks reveal that they are working on Stubbs, a new feature that allows users to create and deploy AI-generated app prototypes directly from Makersuite. This development could significantly boost the app industry, creating AI-generated app prototypes similar to Figma. However, it seems unlikely that the Stubbs feature will be able to generate the code files for the app - instead it will focus on the protype’s UI.
3. Amazon unveils new GenAI tool for advertisers [Link]
The new AI tool will allow advertisers to create themed backgrounds for their product images - simply upload a photo of the product, describe the background you want, and select a theme. Amazon have introduced other tools that focus on product descriptions and summarising customer reviews.
4. Google announce new feature to fight against misinformation [Link]
The company is now rolling out the “About this image” feature to provide more context about images to combat misinformation, including whether it was made using AI. Google users will be able to view an image’s history, metadata, how it was used on various sites and when it was first detected by Google Search.
This will be an invaluable tool, as images that are several years old are often re-used to spread misinformation on social media. Just open Google Images, click on the image, and then the three-dot menu.
5. Cruise suspend all their robotaxi operations, after California agencies revoked their permits [Link]
Last week we outlined why the self-driving company Cruise are under the microscope in the US and that has started to intensify in recent days. California’s DMV have suspended Cruise’s robotaxi permits, after they alleged that the company withheld video footage from them. Cruise dispute this but the company has since decided to suspend all their operations in Austin, Houston, and Phoenix - with the aim to “earn public trust” again.
Announcement
UN creates a new AI advisory body for international regulation
This follows on from the UN’s gathering of AI experts just a few weeks ago. Gary Marcus, who’s a leading figure in the calls for AI regulation, later said that the new body will have “significant powers” - it’ll be interesting to see what the UN recommends.
The new body includes figures from OpenAI, Microsoft, Sony, Stanford, and university professors from around the world. Their aim is to recommend how AI can be regulated internationally, in the face of rapid technical advances, and to minimise the harms of AI while also maximising the benefits. The UN’s Secretary-General is concerned that it could be used to “undermine trust in institutions, weaken social cohesion and threaten democracy itself”.
It’s expected their interim report will be published by the end of this year, with the final version completed in the middle of 2024. There are also plans to hold an AI summit next September to discuss the way forward on regulation.
You can look at who’s sitting on the committee here.
Byte-Sized Extras
X wants to replace your bank account [Link]
Google invests $2 billion in Anthropic AI [Link]
LAION, group behind Stable Diffusion, want to open source AI systems that can recognise emotion [Link]
Magic Leap, the unprofitable Augmented Reality company, changes their CEO just months before Apple enters the sector [Link]
Google pays $26 billion to be the world’s default search engine, every year. Most of it goes directly to Apple [Link]
Startup Spotlight
Viso
Viso are an interesting start-up, with their own low-code platform used to quickly deploy computer vision (CV) applications. Instead of having to build CV apps from scratch, companies can either use Viso’s off-the-shelf models or customise them for their specific use case.
The entire process can be managed within the platform - such as data annotation, training, deployment and monitoring - and often with a drag-and-drop interface. They’ve recently secured $9.2 million in funding and some of their customers include PwC, DHL, and DPD.
If you want to read more about what Viso do, have a look at their website.
Analysis
This has been a terrible week for Cruise. Their CEO, Kyle Vogt, was full of optimism a few weeks ago at the TechCrunch Disrupt conference - but this does not look optimistic for the company. While Cruise have stopped all their robotaxi operations, their competitors are able to continue collecting data. In recent months, it has been Cruise who have faced the most news attention and criticism - to the benefit of other self-driving companies, such as Waymo.
Last week, I flagged that concerns around Cruise’s safety record could snowball into something much bigger. Unfortunately, that looks to be the case. If they’re not careful, the company could face more opposition that will try to block their expansion into other US cities.
This is the last thing Cruise need, as they have plans to expand into the Japanese market in the coming years. But if governments are worried that Cruise can’t be trusted to supply them with the complete data after a crash, then that’s all a very distant dream.
Some good news came from Google with the release of a new feature to tackle misinformation. Users can now see details about when an image was first detected by Google and whether or not it may have been generated using AI. It’s a common issue on social media, with images often re-used to spread a particular message.
We’ve seen some of that happen in recent weeks, as war images from 10 years ago have been recycled by bots online. Of course, this feature doesn’t stop misinformation from being spread, but it does allow people to get more context as to when an image was first posted online and whether it’s being misrepresented by malicious actors.
This Week’s Art
Prompt: 8k high resolution hyper-realistic digital painting of a western scene. A stagecoach is seen in the distance, passing between towering snowy mountains under a vast night sky dominated by a luminous moon. The scene is bathed in a mix of bright and dark colors, adding depth and contrast. In the forefront, a cowboy sits next to a blazing campfire, cooking food. The detailed 3D render captures every nuance, from the textures of the cowboy's attire to the snow on the mountains, creating a cinematic masterpiece.
Platform: DALL-E 3
End Note
It’s been another interesting week, with several new GenAI tools being worked on by both Amazon and Google. As those companies start to build the scaffolding needed to get reliable results from LLMs, we’re starting to see more product launches. But with the fierce competiton from Microsoft, they need to keep them coming.
We’ve also looked at the UN’s new AI body for international regulation, Cruise’s struggles, and the doomsday team that’s been formed within OpenAI - along with Viso’s computer vision platform for quick app deployment. It’ll be interesting to see what comes from the UK’s AI Safety Summit later on this week.
Have a good one!
Liam
Share with Others
If you found something interesting, feel free to share this newsletter with your colleagues.