Main AI Updates

GPT-4o Advanced Voice Launching This Month

OpenAI CEO Sam Altman says the first users will start to get access to GPT-4o Advanced Voice in the next couple of weeks, but this will be a limited "alpha" rollout.

Key Features

GPT-4o Advanced Voice is an entirely new type of voice assistant that can create custom character voices, generate sound effects while telling a story, and even act as a live translator.

Hold-up and Safety Measures

OpenAI is concerned that GPT-4o Advanced Voice, without appropriate guardrails, could offer potentially harmful information or be used unexpectedly.

To tackle this, they will launch the Alpha with a small group of users to gather feedback and expand based on what they learn.

Future Updates

Future updates will add live vision features, letting the AI see what you see, enhancing its interaction capabilities and making it an even more versatile assistant.

⚡️ OpenAI Introduces GPT-4o Mini

AI Accessibility: Affordable Multimodal Model

OpenAI just dropped GPT-4o Mini, a compact and budget-friendly twist on their powerful GPT-4o model.

Performance and Cost

GPT-4o Mini is a fully multimodal model currently capable of generating text and images.

It scored a solid 82% on the MMLU benchmark, surpassing GPT-3.5 Turbo (70%) and challenging Claude 3 Haiku (75.2%) and Gemini 1.5 Flash (78.9%).

Remarkably, it’s up to 60% cheaper than GPT-3.5 Turbo, making it a game-changer for developers on a tight budget.

Capabilities

It outshines GPT-3.5 and holds its own against GPT-4 in many tasks, handling 128,000 tokens effortlessly for complex interactions and extensive data processing.

This model opens new possibilities for developers looking for high performance at an affordable cost.

🚫 AI Trained on YouTube Without Consent

Ethical Concerns: Unauthorized Data Use

A new investigation by Proof News has revealed that tech giants, including Apple, Anthropic, Nvidia, and Salesforce, used content from over 170,000 YouTube videos to train their AI models without creators’ consent.

Details of the Dataset

The dataset, called “YouTube Subtitles,” contains transcripts from over 48,000 channels, including popular creators, news outlets, and learning channels.

Nonprofit EleutherAI compiled the data as part of a larger collection called ‘The Pile,’ intended to provide training materials for developers and academics.

Creator Unawareness

Creators were unaware that their content had been used for AI training purposes, and YouTube’s Terms of Service also prohibit the use without permission.

Apple reportedly used the dataset to train OpenELM, a model related to new AI features for iPhones and MacBooks.

Implications

While the use of these transcripts might not lead to legal ramifications for the firms involved, it certainly raises ethical and moral concerns.

The report highlights the ongoing issues with unauthorized data use in AI training, and its potential impact on creators and the broader AI community.

Latest Developments

🖼️ Bring Your Images to Life with Motion

Creative Tools: Leonardo AI’s Motion Feature

Leonardo AI has introduced a new ‘Motion’ feature that allows users to turn static images into captivating short animations.

This feature is designed for social media, web design, digital artists, and more.

How to Use Motion

1. Sign up on Leonardo AI’s website (a free account includes 150 daily credits).

2. From the main dashboard, click on "Image Generation" in the sidebar menu.

3. Generate an image using the prompt of your choice.

4. Pick your favorite image, hover over it, and click the ‘Motion’ button.

5. Adjust the Motion Strength slider as desired.

6. Click "Generate" and check out your animated creation!

This new feature provides an exciting way for users to enhance their static images with dynamic elements, making their content more engaging.

👀 OpenAI Trains AI to Explain Itself Better

Research Breakthrough: Verifiable AI Outputs

OpenAI has published new research detailing a method to make large language models produce more understandable and verifiable outputs.

This technique involves a game played between two AIs to make generations more ‘legible’ to humans.

Prover-Verifier Game

The method uses a "Prover-Verifier Game" where a stronger AI model (the prover) tries to convince a weaker model (the verifier) that its answers are correct.

Through multiple rounds of the game, the prover learns to generate solutions that are not only correct but also easier to verify.

Results and Applications

While the method boosted accuracy by about 50% compared to optimizing solely for correctness, its solutions were easily checkable by humans.

Significance

This research offers a scalable way to potentially keep systems ‘honest,’ though the performance trade-off highlights the challenge of balancing capability with explainability.

🤯 New Sora Competitor: Haiper 1.5

High-Quality Video Generation

Another competitor in the video generation space, Sora Al, has just released Haiper 1.5, a tool that can generate up to 8 seconds of high-quality video and includes an upscaler for HD generations.

User Experience

Users can try the tool for free at haiper(dot)ai.

Here's an example prompt to test the tool:

"POV of a man on a moving train with reflections of the Swiss Alps on the train window."

This tool offers new possibilities for creating engaging video content with ease.

📲 Microsoft’s AI Designer App Goes Mobile

Tech Innovation: AI-Powered Design

Microsoft has announced that its AI-powered Designer app is now available worldwide, bringing advanced image generation, editing, and design capabilities to users on mobile and Windows platforms. 

This marks a significant step in making AI design tools accessible to a broader audience.

Key Features

Designer is available in over 80 languages on the web, on Android and iOS, and via Windows platforms.

The app uses AI to generate images and designs from text prompts, allowing users to create custom stickers, emojis, avatars, and more.

New Capabilities

New features include ‘prompt templates’ for fast creation, ‘Restyle’ for remixing existing images, and ‘Frame’ for creating personalized frames and collages.

Microsoft's push into the AI design space comes amid competition from major players like Canva and Adobe, signaling a transformation in how people approach design.

Top AI Stories

🤖AI Stories

🦾 Anthropic is reportedly working on a new screenshot tool for Claude, potentially allowing users to seamlessly take screenshots from other screens or tabs.

📜 California state legislators are pushing for a proposed bill that would require big tech companies to test AI for "catastrophic" risks before public release.

💻 AMD claims its new laptop chips can outperform Apple's M3 — boasting improved performance in multitasking, image processing, and gaming.

🖼️ Google is reportedly developing a ‘Prompts Gallery‘ to allow users to curate a collection of favorite prompts and get inspiration from other users within it’s Gemini chatbots.

🗣️ Perplexity rolled out Voice Mode to it’s AI assistant on its iOS app, allowing Pro users to chat and ask questions to the AI search engine through various voice mode

Thanks for reading!

If you enjoyed this, please help spread the love by forwarding this Newsletter to a friend or colleague.

SPONSOR US

Get your product in front of over 4000+ AI enthusiasts

Our newsletter is read by thousands of tech professionals, investors, engineers, managers, and business owners around the world. Get in touch today.

FEEDBACK

I hope to see you in the next one!

AI Genesis