New Anthropic AI Uses Computers Like Humans Do
Plus: Ex-OpenAI researcher alleges copyright breach, DeepMind open-sources its AI watermarking tool, Midjourney’s new AI tool lets you edit any web image, and more.
Hello Engineering Leaders and AI Enthusiasts!
This newsletter brings you the latest AI updates in just 4 minutes! Dive in for a quick summary of everything important that happened in AI over the last week.
And a huge shoutout to our amazing readers. We appreciate you😊
In today’s edition:
🖥️ Anthropic’s new AI controls computers like humans do
🚨 Ex-OpenAI researcher alleges copyright breach
📝 DeepMind open-sources SynthID, its AI watermarking tool
🖌️ Midjourney’s new AI tool lets you edit any web image
🤖 OpenAI dissolves AGI Readiness team
📱 Quantized Llama 3.2: 56% smaller, 4x faster on mobile
📚 Knowledge Nugget: AI Integration or Generative Education? by
Let’s go!
Anthropic’s new AI controls computers like humans do
Anthropic has announced that its AI model, Claude, can now control computers like a human, navigating screens, clicking buttons, and typing text. This "computer use" capability could revolutionize automation in industries that rely on repetitive tasks across multiple applications.
Anthropic has also released upgraded versions of its AI models, with the Claude 3.5 Sonnet showing significant improvements in coding and tool use and the new Claude 3.5 Haiku matching the capabilities of previous high-end models at a lower cost and faster speed.
Why does it matter?
Anthropic's Sonnet and Haiku updates are surprisingly strong. By adding computer abilities to these models, Anthropic is challenging many automation startups, as these could transform how businesses automate complex workflows.
Ex-OpenAI researcher alleges copyright breach
Suchir Balaji, a former OpenAI researcher who left in August, has spoken out against the company's use of copyrighted materials to train its AI models. He claims that OpenAI's practices violate copyright law by making copies of training data, which harms original creators.
Balaji worked on GPT-4 training and argues that the company's use of online data should not qualify as 'fair use.' These allegations come amidst OpenAI’s constant churn of high-profile departures.
Why does it matter?
While many experts warn about AI’s future risks, this insider perspective spotlights pressing legal and ethical concerns in today’s AI development. Even pioneers like OpenAI may be pushing boundaries at the expense of copyright norms, which could erode trust in the AI sector.
DeepMind open-sources SynthID, its AI watermarking tool
Google DeepMind has open-sourced its SynthID technology, a tool for watermarking AI-generated content to help identify its source. SynthID embeds an invisible digital watermark into text, images, audio, and video without compromising quality, accuracy, or creativity.
The watermarking process works by adjusting the probability scores of the tokens generated by language models, creating a unique pattern that the system can detect. SynthID has already been integrated into Google's AI products, like Gemini, ImageFX, VideoFX, and Vertex AI.
Why does it matter?
As concerns over misinformation and copyright infringement rise, democratizing watermarking could enhance transparency and accountability in AI-generated content, which is crucial as AI continues to permeate various sectors.
Midjourney’s new AI tool lets you edit any web image
Midjourney has launched a new web-based editor that allows users to modify existing images easily. The tool offers features like retexturing, expanding, cropping, and stylizing images using text prompts and also works with existing features like personalization and style references.
Another new re-texturing tool lets users adjust lighting, materials, and textures while preserving image structure. Access is initially limited to yearly subscribers, 12-month members, and users with 10,000+ generations.
Why does it matter?
The ability to edit authentic web images fundamentally challenges digital authenticity and copyright. It blurs the line between original and modified content, potentially undermining photographers' work and complicating online image ownership and consent questions.
OpenAI dissolves AGI Readiness team
OpenAI has disbanded its AGI Readiness team, a group that advised the company on handling increasingly powerful AI systems. The team will be moving under new Chief Economist Ronnie Chatterji, with the rest distributed across other teams.
This comes after the departure of a senior advisor, Miles Brundage, who expressed concerns about the industry's lack of preparedness. He said in his resignation that neither OpenAI nor any other lab is ready for advanced AI, and the world is also unprepared.
Why does it matter?
This follows the earlier disbandment of OpenAI’s Superalignment team, suggesting a potential deprioritization of long-term AI safety and alignment. It deepens concerns about oversight, as rapid AI advancements may happen with fewer dedicated internal safeguards.
Quantized Llama 3.2: 56% smaller, 4x faster on mobile devices
Meta has released quantized versions of its Llama 3.2 language models, designed to run efficiently on mobile devices. These new models are up to 4 times faster in inference speed, 56% smaller in size, and use 41% less memory compared to the original BF16 format.
Meta achieved these improvements through two fundamental techniques: Quantization-Aware Training with LoRA adaptors and a state-of-the-art post-training quantization method called SpinQuant. The quantized Llama models maintain quality while improving performance.
Why does it matter?
These optimized models could enable more efficient on-device AI computing, allowing developers to build lightweight, high-performing applications that run directly on mobile phones and other resource-constrained devices.
Enjoying the latest AI updates?
Refer your pals to subscribe to our newsletter and get exclusive access to 400+ game-changing AI tools.
When you use the referral link above or the “Share” button on any post, you'll get the credit for any new subscribers. All you need to do is send the link via text or email or share it on social media with friends.
Knowledge Nugget: AI Integration or Generative Education?
In this post,
argues that higher education should focus on "generative education" to shape students as thinkers and creators rather than just job training. He says integrating AI throughout the curriculum is misguided, as it undermines the formative process that allows students to experience intellectual activities like reading, writing, and critical thinking. The author advocates a mindset shift from job preparation to a holistic, student-centered approach that prioritizes personal growth and development.Why does it matter?
Making the education system heavily AI-focused could create a generation dependent on tools rather than developing fundamental cognitive abilities and creative thinking skills through direct intellectual engagement
What Else Is Happening❗
🤖Google is developing Project Jarvis, an AI assistant that can control users' web browsers to automate tasks like booking flights or buying tickets.
🚀OpenAI's new sCM approach generates high-quality samples faster than diffusion models, opening up possibilities for real-time image, audio, and video generation.
📰Microsoft and OpenAI are giving news outlets $10 million in grants to hire AI fellows and explore using AI tools for journalism tasks.
🎶DeepMind has unveiled new AI-powered music creation tools, including MusicFX DJ for interactive music generation and updates to its Music AI Sandbox and YT’s Dream Track.
🎙️ElevenLabs introduces Voice Design, an AI feature that generates a unique, customizable voice from a simple text prompt.
🚗Qualcomm and Google are partnering to help car companies create custom AI voice assistants for vehicles using Qualcomm hardware and Google's Android Automotive OS.
🎨Canva has added new AI features, including a text-to-image generator called "Dream Lab" that uses its recent acquisition of Leonardo.ai.
🎥Genmo launched Mochi 1, an open-source AI video generation model that claims to rival leading closed-source competitors like Runway and Kling.
🗞️Meta will use Reuters news content to train its AI chatbot, which will provide news and information to users on Facebook and Instagram.
📚Goodreads co-founder launched an AI-powered app called Smashing that curates web content and allows users to engage with stories from different perspectives.
New to the newsletter?
The AI Edge keeps engineering leaders & AI enthusiasts like you on the cutting edge of AI. From machine learning to ChatGPT to generative AI and large language models, we break down the latest AI developments and how you can apply them in your work.
Thanks for reading, and see you next week! 😊