- The Daily Update
- Posts
- Apple eyes on-device AI
Apple eyes on-device AI
Insight: Image generators are trained on explicit photos of children
Jack Shields
December 21, 2023
Welcome back!
It’s one of those days again — some exciting news from Apple is being overshadowed by a horrifying new study that’s already causing controversy in the world of AI.
Let’s jump in.
In today’s Daily Update:
🗞️ Apple prepares for on-device AI
🤖 AI image generators are being trained on explicit photos of children
📸 Microsoft Copilot can now generate full songs
🚨 AI Roundup: Four quick hits
Read time: 2 minutes
TOP STORY
🗞️ Apple prepares for on-device AI
Generated by Adobe Firefly
Apple just published a new paper that outlines a method for running AI models on devices.
What you should know:
The company’s method allows devices to run LLMs that surpass their DRAM capacity.
It relies on a flash memory framework that optimizes the transfer and reading of data.
This approach enables devices to run models up to twice the size of the available DRAM with impressive speed.
Apple plans to integrate generative AI capabilities into iOS 18.
Why it matters: It won’t be long before Apple incorporates AI models into all of its devices. This will give over 1.46 billion iPhone users access to on-device AI for the first time.
RESEARCH INSIGHT
🤖 AI image generators are being trained on explicit photos of children
Screenshot: Disabled LAION AI training datasets
A new study by Stanford researchers reveals that training datasets for popular AI image generators include thousands of explicit photos of children.
The details:
The Stanford Internet Observatory found more than 3,200 images of suspected child sexual abuse in the LAION database, which has been used to train leading text-to-image models such as Stable Diffusion.
Researchers previously thought that AI models were only able to generate explicit photos of children by combining adult pornography with benign photos of kids.
LAION temporarily removed its datasets prior to the release of the report.
Explicit AI-generated photos of children have already been disseminated in schools around the world.
Hard truth: LAION and AI startups are working urgently to remove illegal content from their training datasets. Unfortunately, older versions of AI image generators are already embedded in other applications and tools. This means that many people will always have access to models capable of producing explicit imagery of children.
BUSINESS SPOTLIGHT
📸 Microsoft Copilot can now generate full songs
Source: Microsoft
Microsoft has integrated the music creation app Suno into its Copilot conversational AI. This allows users to turn text prompts into AI-generated tracks.
Key points:
Suno uses Copilot’s language capabilities to synthesize songs that include lyrics, instrumentals and vocals.
The new feature is rolling out to Microsoft Edge users and is available via the Copilot plugins tab.
Suno says its models don’t recognize artists by name and block certain prompts.
The relevance: AI-generated music is a hot topic among artists and record labels striving to protect their intellectual property. Microsoft’s new tool will raise even more questions about the future of the traditional music industry.
MORE TRENDING NEWS
🚨 AI Roundup: Four quick hits
Source: dictionary.com
AI Effect: Dictionary.com selects “hallucinate” as its 2023 Word of the Year.
Medical AI: MIT researchers identify a new class of antibiotic candidates.
Access Denied: RiteAid is banned from using facial recognition software after falsely identifying shoplifters.
Not so Smart Salesman: Users trick ChatGPT into selling a car for $1.
THAT’S ALL FOR TODAY