- Starter AI
- Posts
- Meta’s “Imagined with AI”, Smaug-72B breaks records, Apple’s new AI model
Meta’s “Imagined with AI”, Smaug-72B breaks records, Apple’s new AI model
There’s a label for everything, especially in the AI era.
Sthefania
February 07, 2024
Hello, Starters!
We’ve got a few models to cover, the open source community is always filled with interesting advancements, and here in StarterAI, we’re all set to help you stay updated.
Here’s what you’ll find today:
Meta introduces “Imagined with AI”
Smaug-72B is the next big thing in open source
Apple unveils “MGIE”
OpenAI announces watermarks for DALL-E 3 content
OnePlus gets into the AI smartphone trend
And more.
🖼️ Meta introduces “Imagined with AI” (2 min)
In order to fight deep fakes and promote more transparent and responsible AI development, Meta has announced plans to watermark AI-generated content with a visible label, "Imagined with AI," and include an invisible mark along with the corresponding metadata.
Meta is not the only company pursuing this effort; major industry names such as Microsoft, Adobe, and Midjourney are also taking steps to label their content, as the surge of misleading AI-generated content raises concerns for users.
If you're looking for an interesting match to keep track of, make sure to head to Hugging Face's open LLM leaderboard. There's always a new contender trying to break records with its impressive benchmarks, and the current main character is Smaug-72B, surpassing GPT-3.5 and Mistral Medium.
This model, introduced by Abacus AI, aims to help enterprises solve complex problems involving AI and machine learning. It's worth noting that Smaug-72B is a fine-tuned version of Qwen-72B, a powerful model released by Qwen, a team of researchers at Alibaba.
🍎 Apple unveils “MGIE” (4 min)
As the weeks go by, we have found even more evidence of Apple's slow and steady journey in AI. The recent release involves an open source AI model dubbed "MGIE," developed in collaboration with researchers from the University of California.
MGIE leverages multimodal large language models (MLLMs) to interpret users' prompts and perform pixel-level manipulations. These kinds of models are efficient enough to process both text and images, enabling instruction-based image editing. The research could potentially hint at Apple's upcoming AI announcements at this year's WWDC.
🖼️ Following Meta's footsteps, OpenAI has announced that it's adding watermarks to DALL-E 3 generated images. Watermarks from the Coalition for Content Provenance and Authenticity (C2PA) will be added to the metadata to ensure trust between users and ensure safety.
🕹️ OnePlus has decided to catch up with the AI race on smartphones, although its OnePlus 12 didn't have any AI capabilities during its launch. An upcoming software update will bring an "AI Summarizer" to recap phone calls, a photo editor, and a feature to create summaries of articles.
Sponsored content:
Your info is on the dark web
While other personal data removal services only focus on one type of data broker, Incogni helps remove your personal information from all broker types, including your information visible on People Search Sites (e.g. WhitePages.com). Our readers exclusively get 55% off Incogni annual plans with code PRIVACY.
⚡️Quick links
What did you think of today's newsletter? |