- The AI Digest
- Posts
- Superintelligence may be here sooner than expected!
Superintelligence may be here sooner than expected!
Jason - The AI Digest
September 30, 2024
Welcome, AI Enthousiasts!
In today’s AI newsletter:
→ AI Newsflash
→ Meta unveils AR xAI glasses, new model, and more
→ OpenAI CTO exits amid rumors of non-profit removal
→ OpenAI rolls out Advanced Voice Mode
→ Google releases production-ready models
→ Superintelligence may be here sooner than expected
→ Nvidia and Alibaba collaborate to improve autonomous cars
→ 5 Best AI tools for podcasting
Reading time: 9 minutes
AI Newsflash
OpenAI is reportedly working on an enhanced version of its Sora AI video generation model, aiming to deliver higher quality and longer video clips compared to previous iterations.
Apple announced gradual AI-powered updates to Siri, with the most significant improvements expected in iOS 18.3, set to launch by January 2025.
OpenAI (Advanced Voice Mode) CEO Sam Altman confirmed the early completion of the Advanced Voice Mode rollout for ChatGPT, except in areas requiring additional external approval.
Duolingo introduced AI-powered features, including Adventures mini-games and a Video Call tool, designed to provide more immersive and practical language-learning experiences.
OpenAI (API Access) expanded API access for its o1 models, adding tier 4 with 100 requests per minute and increasing tier 5 users' limit to 1000 requests per minute.
Warner Bros. Discovery adopted Google Cloud’s AI to generate captions, aiming to reduce production time and costs for unscripted television programming.
LinkedIn suspended its AI training on UK users’ data following privacy concerns raised by the Information Commissioner’s Office, pausing the practice pending further discussion.
During its Connect 2024 event, Meta unveiled a range of cutting-edge AI innovations, including the introduction of its Orion AR glasses, the Llama 3.2 model, and new AI tools for Reels. Major enhancements to Meta AI were also showcased, including a Voice interaction mode.
Key highlights:
Meta showcased its Orion AR glasses prototype, featuring a lightweight design (under 100g), a broad field of view, and advanced capabilities like voice commands and hand tracking. The development of these glasses has taken over a decade.
The company also revealed Llama 3.2, a powerful vision model that can interpret both text and images. This model comes in 11B and 90B parameter versions.
Additionally, smaller Llama models with 1B and 3B parameters were announced, designed for efficient use in devices like smartphones and potentially future wearables.
Instagram is set to receive AI upgrades, including automatic dubbing and lip-syncing in various languages for creators, alongside personalized AI-generated content recommendations in the feed.
Meta also introduced Voice Mode, enabling users to interact with Meta AI using their voice on platforms like Messenger, Facebook, WhatsApp, and Instagram DMs, similar to ChatGPT’s voice feature.
Meta Connect 2024 marks a significant moment for the company, with the introduction of open-source models, innovative AR technology, and the integration of AI voice features across its platforms. With nearly half a billion active users interacting with Meta AI, the company continues to push the boundaries of tech innovation, proving its leadership in the field.
Mira Murati, who has served as OpenAI's CTO for over six years, has announced her departure from the company. This move comes amidst speculation that OpenAI is transitioning away from its non-profit model, possibly granting CEO Sam Altman equity in the company.
Key points:
Murati has cited the desire for personal exploration as the reason for her departure, while also ensuring a smooth transition during this period.
Her exit coincides with reports suggesting that OpenAI is restructuring its business to become a for-profit benefit corporation. In this new structure, Sam Altman is expected to receive equity, with the non-profit side of OpenAI maintaining only a minority stake.
During her farewell, Murati expressed her gratitude for her tenure at OpenAI, mentioning key milestones like the development of speech-to-speech technology and the OpenAI o1 project.
CEO Sam Altman responded by commending Murati’s impact on the company and indicated that further details about the transition will be shared soon.
This leadership change adds to a series of high-profile exits at OpenAI, following the departure of Andrej Karpathy in February, Ilya Sutskever in May, and Greg Brockman’s recent sabbatical in August. With Sam Altman being the sole remaining top leader, many are left wondering about OpenAI’s future direction—though answers remain elusive, with only a vague promise of updates "soon."
OpenAI is expanding its Advanced Voice Mode (AVM) to all ChatGPT Plus and Teams users this week, introducing enhanced voices and improved features to create more natural and personalized AI interactions.
Key points:
The initial version of Advanced Voice Mode was launched in July but was only available to a limited number of users.
Since then, OpenAI has enhanced AVM with Custom Instructions and Memory, enabling more tailored conversations and better recall of previous interactions.
Improvements have also been made in the system’s ability to understand various accents, offering smoother, faster dialogue. Additionally, five new nature-inspired voices were added, while the "Sky" voice, previously likened to Scarlett Johansson, has been removed.
However, AVM will not be accessible in some regions, including the EU, UK, Switzerland, Iceland, Norway, and Liechtenstein.
As discussions around AI agents and superintelligence gain traction, the relevance of advanced AI interactions becomes increasingly critical. OpenAI’s AVM is designed to make conversations with AI more human-like, addressing a key need for everyday interactions with AI-powered systems.
Google has introduced major upgrades to its Gemini AI models, enhancing performance, cutting costs, and expanding accessibility for developers.
Key highlights:
Two new models, Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002, are now available, providing notable quality improvements, including a 20% enhancement in math-related benchmarks.
Google has reduced the cost of using Gemini 1.5 Pro by over 50% for input and output on prompts under 128K tokens, while also raising rate limits for more usage flexibility.
These updated models deliver results twice as fast and reduce latency by three times compared to previous versions. They also bring better long-context understanding and enhanced vision capabilities.
Developers now have more control over model configurations with updated filter settings, allowing for better customization based on specific needs.
Google's swift iteration on the Gemini models is making AI development more accessible and affordable, empowering developers to create faster, more efficient, and cost-effective applications. Though not yet Gemini 2, these improvements mark a significant step forward, enabling smarter solutions with enhanced performance.
OpenAI CEO Sam Altman has recently speculated that superintelligent AI could emerge within just a few thousand days, signaling a pivotal moment in human history that could unlock immense potential for progress and innovation.
Key points:
Altman foresees AI equipping humanity with tools to address complex challenges, accelerating progress in ways we’ve never experienced before.
He predicts the rise of personal AI teams composed of virtual experts across various fields, capable of helping us create nearly anything we can imagine.
Potential applications could include individualized AI tutors, advancements in healthcare, and the ability to develop any type of software on demand.
Altman stressed the importance of abundant computational resources and energy to ensure AI's broad accessibility, potentially ushering in what he calls an "Intelligence Age," marked by human prosperity and major scientific breakthroughs.
This research highlights how mimicking human cognitive strategies can advance AI performance. The simplicity of RE2 suggests that many AI companies may be missing out on basic, human-inspired approaches while chasing more complex solutions.
NVIDIA & ALIBABA
Nvidia and Alibaba collaborate to improve autonomous cars
Alibaba Cloud and Nvidia have announced a strategic partnership to develop cutting-edge AI solutions for autonomous driving, combining Alibaba's large language models with Nvidia’s automotive computing platform.
Key points:
Alibaba's sophisticated Qwen AI models will be integrated into Nvidia's Drive AGX Orin platform, currently used by major Chinese electric vehicle manufacturers.
This collaboration aims to improve in-car voice assistants, enabling more natural conversations and smarter recommendations by analyzing visual and environmental data.
In addition, both companies are adapting Alibaba’s AI models for Nvidia's upcoming Drive Thor platform, which will integrate advanced driver assistance, autonomous driving features, and AI-powered driver capabilities.
This collaboration between two AI giants promises significant advancements in the automotive sector, particularly in enhancing autonomous driving technology. Nvidia’s decision to incorporate Alibaba's Qwen models is also a noteworthy win for open-source AI, further pushing the boundaries of innovation in the field.
5 Best AI tools for podcasting
LaunchPod - An AI-driven tool designed to automate the entire podcast production process, simplifying content creation from start to finish.
CastMagic - Uses AI to transcribe and summarize podcast episodes, making post-production workflows more efficient.
Descript - Offers AI-powered tools for audio and video editing, allowing podcast creators to transcribe and edit through text-based commands.
Podcastle - A platform that leverages AI to record, edit, and enhance podcast audio, improving sound quality for creators.
Soundraw - An AI tool that generates custom music, enabling podcasters to create personalized soundtracks and background music.
That’s a wrap!
We had a lot to talk about, so let’s wrap it up. If you have any questions, feel free to shoot over an e-mail and we wil get you a response within 24 hours.
If you have specific feedback or anything interesting you’d like to share, please let us know by replying to this email.