Daily AI Newsletter: OpenAI releases ChatGPT’s hyper-realistic voice

In today’s email:

🤕 Making AI models ‘forget’ undesirable data hurts their performance
🔥 Meta is rolling out its AI Studio in the US for creators to build AI chatbots
👯 Friend’s $99 necklace uses AI to help combat loneliness
🧰 11 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

OpenAI releases ChatGPT’s hyper-realistic voice to some paying users

OpenAI recently introduced ChatGPT Advanced Voice Mode, initially available to a select group of ChatGPT Plus users. This new feature, based on the GPT-4o model, promises hyper-realistic audio responses and significantly lower latency in voice interactions by integrating voice-to-text and text-to-voice processes within a single model. OpenAI showcased a prototype voice named "Sky" resembling Scarlett Johansson in May but later withdrew it following legal concerns raised by the actress. The feature is expected to expand to all Plus users by fall 2024.

The Advanced Voice Mode differs from earlier versions by its ability to process emotional nuances in speech and by operating without separate models for different tasks. OpenAI plans a careful rollout to monitor the tool's use and has prepared four preset voices in collaboration with paid voice actors, ensuring the voices don’t mimic specific individuals or public figures. This approach is in response to potential deepfake controversies and legal challenges that have surfaced in the AI industry.

In addition to technical advancements, OpenAI has introduced new safeguards, such as filters to prevent the generation of copyrighted audio content. These measures come as the AI field faces increasing scrutiny and legal actions from entities like record labels. A comprehensive safety report on these new features is scheduled for release in early August.

Use AI As Your Personal Assistant

Ready to embrace a new era of task delegation?

HubSpot’s highly anticipated AI Task Delegation Playbook is your key to supercharging your productivity and saving precious time.

Learn how to integrate AI technology into your processes, allowing you to optimize resource allocation and maximize output with precision and ease.

Download Now!

Introducing Meta Segment Anything Model 2 (SAM 2)

Meta, led by CEO Mark Zuckerberg, has launched Segment Anything 2 (SA2), an advanced vision AI model that extends its segmentation capabilities from still images to video. This was showcased at the SIGGRAPH event where Zuckerberg discussed the model with Nvidia CEO Jensen Huang. SA2 enables more efficient video processing without overwhelming computational resources, making it a significant leap forward from its predecessor which was limited to individual image frames.

The new model is part of Meta's commitment to open AI, providing the SA2 software free for public use, along with a large database of 50,000 annotated videos specifically created for training it. An additional collection of over 100,000 videos, used internally for further training, remains undisclosed, leading to speculation about its sources and availability. This approach underscores Meta's strategy to foster an ecosystem around their AI technologies, which enhances their functionality and broadens their applicability.

Zuckerberg highlighted the practicality and necessity of open-sourcing tools like SA2, not solely driven by altruism but as a strategic move to optimize the technology's effectiveness. Meta has consistently promoted open AI, from tools like PyTorch to recent models like LLaMa and Segment Anything, setting a high bar for AI performance while also sparking discussions on the true openness of such initiatives. The model's details and its training resources are accessible on GitHub for further exploration and use.

Friend’s $99 necklace uses AI to help combat loneliness

Avi Schiffmann, known for his award-winning COVID-19 tracking website and a Harvard dropout, has unveiled his latest venture—an AI necklace named Friend designed to combat loneliness. Priced at $99 and set to start shipping in January 2025, Friend is a neck-worn device that connects to your smartphone via Bluetooth. It functions by continuously listening and responding to its user, offering features like proactive messages and responses similar to texts, enhancing emotional connection through its physical presence.

Despite the mixed success of similar AI hardware in the startup sphere, Schiffmann has secured $2.5 million in funding for Friend, valuing his company at $50 million. This funding round includes notable investors such as Raymond Tonsing from Caffeinated Capital and founders from Solana. Schiffmann emphasizes that Friend is not intended to serve as a therapist or a productivity tool but purely as an AI companion to address the feelings of loneliness through conversation.

Initially, Schiffmann's project started as a $600 pendant named Tab, designed to track people and transcribe meetings. However, after gathering about $100,000 in preorders last year, Schiffmann shifted focus to Friend, offering Tab's early backers a switch to Friend or a refund. This pivot reflects his commitment to enhancing emotional connectivity through AI, encapsulated in the company’s new "always listening" tagline, with assurances of privacy and control over data.