Must See AI Industry News This Week - 5 October 2024
OpenAI’s DevDay 2024 introduced voice-first AI tools, funding hit $6.6B, and Liquid AI launched efficient models, signaling a shift toward real-time, scalable solutions for cost-effective development.
Most Interesting This Week
OpenAI DevDay 2024: Major Updates
OpenAI's first-ever DevDay introduced exciting features that highlight a shift toward voice-first AI.
The Realtime API enables seamless speech-to-speech interactions, perfect for advanced voice assistants and real-time translation, with support for six preset voices.
Vision Fine-Tuning allows developers to customize GPT-4o’s image understanding with as few as 100 images.
Rolled out Prompt Caching, reducing costs by 50% on repeated prompts within a 1-hour window, and Model Distillation, simplifying the process of training smaller, efficient models using outputs from larger ones.
Together, these updates could boost voice-interactive AI, multimodal solutions, and make AI development more affordable across industries like customer service, healthcare, and education.
OpenAI Raises $6.6 Billion in Record Funding Round
OpenAI closed a $6.6 billion funding round led by Thrive Capital, with backing from major players like Microsoft, Nvidia, and SoftBank, pushing its valuation to $157 billion. This substantial investment shows strong confidence in OpenAI’s long-term potential despite leadership shake-ups and sets the stage for intensifying competition in the AI space, particularly as OpenAI expands its capabilities across voice, multimodal AI, and developer tools.
Liquid AI Launches Liquid Foundation Models (LFMs)
Liquid AI introduced Liquid Foundation Models (LFMs), a series of memory-efficient models designed as a potential alternative to traditional transformers. With models ranging from 1.3B to 40.3B parameters, the smaller LFMs outperform some larger models on benchmarks like MMLU, offering faster performance in on-device AI applications for resource-constrained environments. This development could shift the AI landscape, particularly for industries needing efficient, real-time AI without heavy computational demands.
Quick Bites
California Governor Gavin Newsom vetoed the controversial AI safety bill SB 1047, citing concerns about stifling innovation.
OpenAI is reportedly developing Canvas, a new ChatGPT interface for collaborative writing and coding with AI-powered suggestions.
Google announced ads in AI Overview search summaries and introduced new AI-powered search features like video understanding and voice input for Google Lens.
AI Business Moves
Microsoft Upgrades Copilot with Voice and Vision
Microsoft expanded its Copilot AI assistant, adding voice and vision capabilities that allow users to interact with their content via natural language in Microsoft Edge. This brings Copilot Voice on par with OpenAI’s voice mode, while Copilot Vision enables context-aware assistance within browsers. These updates position Copilot as a key player in the AI assistant market, offering deeper integration into daily productivity tools for both individual and enterprise users.
Stability AI Integrates with Amazon Bedrock
Stability AI announced that its flagship text-to-image models, including Stable Image Ultra and Stable Diffusion 3 Large, are now integrated with Amazon Bedrock. This offers businesses high-speed, scalable image generation directly through Amazon’s cloud infrastructure, further cementing Stability AI’s position as a go-to provider for enterprise-grade image generation tools.
AI In Action
Google NotebookLM Expands to YouTube and Audio
Google’s NotebookLM now supports YouTube and audio files, powered by the multimodal Gemini 1.5 model. Users can transcribe, summarize, and interact with videos and audio, expanding NotebookLM’s utility for content creators, educators, and researchers who rely on multimedia for insights and learning. This feature significantly enhances the tool's application across industries that leverage rich media for knowledge extraction.
MIT’s ‘Future You’ Simulates Aging for Personal Growth
MIT's Future You allows users to interact with an AI simulation of their older selves, using personal data to create age-progressed photos and personalized conversations. Early studies suggest that interacting with future versions of oneself can reduce anxiety and improve self-awareness, highlighting the growing role of AI in mental wellness and personal development.
Black Forest Labs' Flux 1.1 Pro Sets New Benchmark for Image Generation
Flux 1.1 Pro, Black Forest Labs’ latest text-to-image model, generates images six times faster than its predecessor with improved quality and prompt adherence. It has surpassed competitors like MidJourney and DALL-E in performance benchmarks, establishing itself as a leader in efficient, high-quality image generation for industries requiring fast, scalable solutions.
PyTorch's torchao Optimizes Llama 3 for Resource-Constrained Deployment
PyTorch introduced torchao, a new library that reduces inference time by 97% for Llama 3 8B, using advanced techniques like quantization and sparsity without compromising accuracy. By offering faster and more memory-efficient inference, torchao makes deploying large language models (LLMs) feasible for businesses with limited computational resources, opening new avenues for more responsive AI applications.