Malicious code in Open Source, $50 million from Apple to Shutterstock, partial image editing from OpenAI lead the top AI news of the week
Our latest AI Digest covers the biggest breaking AI news of the week. Anywhere Club community leader, Viktar Shalenchanka, comments on key stories.
News #1 — Malicious code in Open Source nearly broke the internet
A dramatic piece of news related to a vulnerability in the liblzma component (part of the XZ package) might not directly correlate to AI, but it certainly sparks some thoughts. Malicious code, introduced by an Open Source SSH contributor, almost slipped into production. This would've allowed attackers to access a large part of the internet — enabling them to disrupt infrastructure, steal information, and plant malware! Fortunately, the vulnerability was noticed by another contributor — purely by coincidence — after the code had entered public test builds. From one perspective, this is a story with a happy ending. From another, we can't help asking: "How many such vulnerabilities have gone unnoticed?" and "What happens when AI agents created by malevolent actors — who introduce their changes into the ’open‘ source while disguising themselves as humans — start to appear in the world?"
News #2 — $50 million from Apple to Shutterstock: awaiting products?
For weeks now, Apple has been making big splashes in the AI news. This time, the company is in the spotlight for its deal with the renowned photobank, Shutterstock, for access to images, video, and audio. The purpose of the transaction is to train AI models. The price tag? US$50 million. Considering the recent news about the ReALM text model, it's evident that we'll see significant AI service integrations in iOS at the next WWDC conference. Apple will finally enter the AI arms race on a product level, not just in news and statements.
News #3 — Partial image editing from OpenAI
OpenAI added a new function for working with images. Now, in the ChatGPT web interface, you can generate images using DALL·E, and also partially edit them. The new technology allows you to select and change details in a small part of the image without changing the entire original picture. This small, yet incredibly significant, function should help content creators work more effectively, especially since partial editing has been the hardest part of working with diffusion models.
Surprise update:
Remember our AI service catalog? It already contains 243 services, all in Open Source! Enjoy!