ElevenLabs AI Phone Agents

ElevenLabs has launched a brand new conversational AI platform that streamlines the creation of custom voice agents, whether for customer support or immersive gaming. With built-in capabilities for speech recognition, text generation, and real-time interruption handling, the platform ensures fluid, human-like conversations. It supports 31 languages and integrates with major AI models resembling Gemini, Claude, and GPT.

Pricing starts at $0.10 per minute, dropping to $0.015 as usage scales. The platform also boasts seamless integration with Twilio, letting businesses plug AI agents directly into their phone systems for automated interactions. Developers get the pliability of SDKs for Python, JavaScript, React, and Swift, together with WebSocket API access. ElevenLabs eliminates the lengthy setup time typically required, offering a faster path to scalable, high-quality AI-driven communication. Read more.

At AWS re:Invent 2024, Amazon rolled out Nova, its latest suite of multimodal AI models aimed squarely on the generative AI space. The family includes Micro, Lite, Pro, and Premier models, each with progressively larger capacities and faster processing times, supporting every part from basic text output to more intricate custom workflows. With token limits starting from 128,000 to 300,000, the models cater to diverse use cases, while Premier offers fine-tuning for bespoke applications.

Nova Canvas and Reel extend the offering, enabling users to generate images and six-second videos with adjustable camera movements. While Reel pushes the envelope with its cinematic capabilities, each tools are anchored by built-in safeguards to make sure responsible content creation. Looking ahead, AWS plans to expand the models with speech-to-speech and “any-to-any” capabilities, further enhancing Nova’s versatility and its positioning within the multimodal AI race. Read more.

Researchers have demonstrated a photonic processor that handles the complete spectrum of deep neural network computations—optically, on a single chip. This bypasses the same old electronic bottlenecks, offering faster, more energy-efficient performance. By leveraging light for each linear and nonlinear operations, the chip eliminates the necessity for off-chip electronics, maintaining a seamless flow of computation.

Achieving over 92% accuracy in under half a nanosecond, it competes directly with traditional hardware. Built with business fabrication techniques, this breakthrough paves the best way for integrating photonics into real-world applications like lidar and high-speed telecommunications.

The challenge now’s scaling this technology and pushing the boundaries of what optics can achieve in real-time machine learning. Read more.