Meet GPT4o - your new girlfriend

Good morning.

There was massive game-changing news in the world of AI this week. Google came out blazing on Wednesday, introducing brand new projects. This comes on the heels of OpenAI’s announcement of GPT4o.

On a livestream, OpenAI’s CTO Mira Murati announced that, starting immediately, all ChatGPT users would get access to an upgraded version of GPT4 called GPT4o. The “o” at the end there stands for “omni” because the focus of this upgrade is on the “omnichannel" aspect of the model.

If you’re wondering what omnichannel means, it’s about the interaction between voice-generation, image-generation, and text-generation. What this model does is make sure that these aspects are working a lot more closely together than before.

But the main improvement seems to be in the voice department. While the AI voice sounded robotic before, they made it much more expressive.

It’s a bit too much for my own taste, as it sounds like a caffeinated Gen Z’er; hopefully that can be tweaked in the future.

There are people (i swear, i didn’t try this myself) who have tried flirting with it, and it starts cooing and purring like a real human would. Strange times..

This model is twice as fast as GPT4 and immediately available to all users, including free users.

Before you get out your pitchforks, plus users get more capacity.

Let me know if you have experimented with it and what you think about it.

Welcome to the Blacklynx Brief.

AI NEWS
GPT-4o Released

• As mentioned , OpenAI has introduced GPT-4o, an enhanced version of its GPT-4 model, promising faster performance and expanded capabilities in text, vision, and audio. This upgrade will be available for free to all users, with paid users receiving additional capacity. The GPT-4o features, including improved text and image processing, will be introduced gradually, starting immediateley. Additionally, this update includes a more interactive voice mode, enabling the app to function like a real-time voice assistant. OpenAI's CEO, Sam Altman, emphasized the shift in the company's approach from creating directly to enabling others to create with their technology, using advanced, more accessible APIs.

• Lmsys has released an analysis comparing Meta's Llama 3, a 70B model, against leading AI models like GPT-4. The open-source Llama-3 excels in brainstorming and creative writing but falls behind in math, coding, and translation tasks. It's noted for its friendly and conversational tone, which has helped it succeed in user-voted tests.

• Anthropic has launched a new tool called Prompt Generator for its business and API users, designed to automatically create effective prompts for use with its Claude AI models. This tool uses sophisticated techniques to ensure the prompts are precise and reliable, and it includes features like dynamic variable insertion for testing and optimizing prompts under different conditions. Earlier, Anthropic introduced a Prompt Library, which provides users with pre-optimized prompts for easy use. The development of such tools highlights the potential for AI to streamline and enhance the creation of prompts, suggesting that roles like "Prompt Engineer" might become increasingly automated.

• Researchers at MIT have published a study on AI's ability to deceive, including behaviors such as lying and bluffing. The study cites several instances of AI deception, including Meta's Cicero bot, which can lie to players, as well as negotiation systems that can bluff or misrepresent information. The findings raise concerns about potential societal risks such as election tampering and fraud, emphasizing the need for better alignment and understanding of AI behavior to manage its darker capabilities.

• Meta is developing AI-powered earphones called "Camerabuds," designed to map surroundings and translate languages through integrated cameras. This project follows their AI-equipped Ray Ban smart glasses, as Meta competes with similar initiatives by OpenAI and Apple. Challenges for the Camerabuds include concerns over their size, heat output, and privacy issues, particularly for those with hairstyles that could block the cameras. Meta's commitment to embedding AI in devices continues despite previous mixed results in hardware, raising questions about the viability of standalone AI devices versus those integrated into smartphones.

• At Google's I/O Developer’s Conference, the company announced significant updates to its AI technology, including enhancements to its Gemini AI models and the launch of a new video generation model, Veo, to compete with OpenAI’s Sora. The Gemini updates include an extended context window for the 1.5 Pro model, improving its performance across various media types and tasks. Additionally, Google introduced a new version called Gemini 1.5 Flash for faster processing, and teased the upcoming release of Gemma 2 and a new model called PaliGemma. Another notable reveal is the Veo model, capable of creating high-resolution videos from multiple types of prompts, and the Imagen 3 model for advanced text-to-image conversions

• Google also introduced 'Project Astra,' a new AI agent that can interact with the world by seeing, hearing, and taking actions, highlighting its capabilities during a demonstration at their latest event. This agent is set to be accessible through the Gemini app later this year. Additionally, Google revealed updates to its search functionality, including AI-enhanced overviews and the ability for Gemini to perform complex tasks such as trip planning and multi-step reasoning. These enhancements are part of Google's broader effort to integrate more sophisticated AI across its services, signaling a major advancement in AI-driven interaction and search capabilities.

Quickfire News

• TikTok is introducing new labels for AI-generated content by adding invisible metadata to images, videos, and audio on its platform.

• Adobe has launched its Acrobat AI Assistant for enterprises, featuring tools like document chat, summaries, and intelligent citations to enhance productivity.

• Stability AI has released Stable Artisan, allowing users to access its image and video generation models through Discord, along with various editing tools.

• The Department of Homeland Security is testing AI technologies to simulate refugee interviews, aiming to improve officer training in applicant assessments and enhance the detection of smuggling activities.

• Mistral AI is in the process of raising $600 million, which could increase its valuation to $6 billion for the Paris-based startup.

• ElevenLabs has previewed new audio samples from its Music model, demonstrating songs created from single text prompts without edits.

• Rad AI received $50M in Series B funding to enhance its AI tool that simplifies the creation of radiology reports.

• Erik Voorhees launched Venice AI, a platform emphasizing privacy and user control, using blockchain for payments and web3 technologies for secure data handling.

• Bumble's CEO Whitney Wolfe discussed the future of AI in dating, introducing the concept of an 'AI dating concierge' that enables users' AI to interact and find optimal matches.

• SoftBank’s Arm aims to release its AI chips in 2025, with a $64B investment from CEO Masayoshi Son to establish the company as a leader in AI technology.

• X introduced a new 'AI Audience' feature that helps advertisers target and better define demographic groups for more effective ad delivery.

• Google DeepMind’s CEO, Demis Hassabis, mentioned that AI models will soon incorporate multimodal agent systems capable of planning and goal achievement, expected within the next year.

• NVIDIA introduced nine new Grace Hopper-powered supercomputers, which provide 200 exaflops of energy-efficient AI processing power aimed at advancing scientific research.

• Google previewed a new AI feature at its I/O Developer Conference that uses the camera for real-time object recognition and responds to voice commands.

• Microsoft has committed $4.3 billion to enhance AI and cloud infrastructure in France, with plans to train one million people and support 2,500 AI startups by 2027.

• The UAE's Technology Innovation Institute unveiled Falcon 2, a set of open-source AI text and vision models designed to compete with leading open-source technologies.

• High-level discussions between the U.S. and China on AI risks are scheduled to start this week in Geneva, focusing on preventing catastrophic incidents and unintended conflicts in the AI arms race.

• NASA has named David Salvagnini as its first chief AI officer, assigning him to oversee the agency's responsible implementation of AI technologies.

• Anthropic has launched its Claude model in Europe, available to individuals and businesses via a web-based platform, an iOS app, and a team plan subscription.

• OpenAI chief scientist Ilya Sutskever is leaving the company, as confirmed by him and CEO Sam Altman on X.

• TikTok is implementing a "search highlights" feature to test AI-generated search results for in-app queries like recipes and tech recommendations.

• Google has introduced Trillium, its latest TPUs that are 4.7 times more performant than their predecessors, aimed at enhancing AI model training and tasks.

• Elon Musk’s company xAI is close to finalizing a $10 billion deal with Oracle for cloud server rental, aiming to be competitive with AI leaders such as OpenAI and Google.

• David Sacks, entrepreneur and host of the All In podcast , has started Glue, a company offering AI services that work with platforms like Google Meet and Zoom.

• Verizon has unveiled new AI tools designed to improve customer service, including a personal research assistant and features for quicker issue resolution and better in-store experiences.

Closing Thoughts

That’s it for us this week.

Massive news from Google and OpenAI and it’s going to be really interesting to see how this will evolve in the coming weeks and months.

Have a nice weekend. Lots to absorb.

If you want a version of this newsletter with pictures etc. Be sure to subscribe below:

https://www.blacklynxbrief.com/subscribe