ElevenLabs Launches Conversational AI 2.0🗣️💎

The world of intelligent agents is undergoing a seismic shift.

ElevenLabs has redefined voice interaction with Conversational AI 2.0 — delivering fluid, multilingual, and secure voice agents built for enterprise scale.

Sakana AI is breaking new ground with its self-improving Darwin-Gödel Machine, an AI that rewrites and evolves its own code to tackle complex programming tasks—no human input required.

And Google AI Edge is putting that power everywhere, offering tools to deploy advanced, multimodal models across mobile, web, and even microcontrollers—keeping data private, apps fast, and AI always-on.

The future of AI agents isn’t coming — it just arrived.

Advanced, Secure, and Multimodal Voice Agents

ElevenLabs has introduced Conversational AI 2.0, a major upgrade to its platform that brings advanced features and robust enterprise capabilities, making it possible to create highly sophisticated and trustworthy voice agents. Compared to the previous version, Conversational AI 2.0 offers a state-of-the-art turn-taking model for more natural, human-like interactions, integrated multilingual support with automatic language detection, and Retrieval-Augmented Generation (RAG) for fast, private knowledge access. The platform now supports multimodal communication (voice, text, or both), batch calling for large-scale outreach, and is built with enterprise needs in mind, offering HIPAA compliance, enhanced security, EU data residency, and high reliability. These improvements enable seamless, intelligent, and secure conversations across industries, empowering businesses to deliver superior customer experiences and streamline operations.

Sakana AI’s Self-Improving Agent

Sakana AI’s Darwin-Gödel Machine (DGM) is an innovative AI system designed to autonomously improve itself by generating, testing, and refining new versions of its own code without human intervention. Drawing inspiration from natural evolution and scientific discovery, DGM explores a wide range of strategies and tools, evaluates their performance on real-world programming benchmarks, and selects the best-performing models to build upon. This iterative self-improvement process has significantly boosted its ability to solve complex programming tasks, increasing success rates by more than double in some benchmarks. Although the system requires substantial computational resources, its ability to evolve independently marks a major breakthrough toward creating AI agents that can continuously learn, adapt, and innovate on their own, potentially transforming how AI development is approached in the future.

Google AI Edge - Model Conversion, Quantization & Optimization

You can now deploy advanced AI models across mobile, web, and embedded devices with tools that let you run the same model on Android, iOS, web, and even microcontrollers—keeping data local for privacy, reducing latency, and enabling offline use. These solutions are cross-platform and multi-framework, supporting models built with JAX, Keras, PyTorch, and TensorFlow. Ready-made, low-code APIs make it easy to add generative AI, vision, text, and audio features to your apps, while flexible frameworks like MediaPipe and LiteRT let you build custom pipelines, visualize and debug model transformations, and accelerate workloads using device-specific hardware like GPUs and NPUs. With tools for model conversion, quantization, and performance benchmarking, you can shorten development cycles and optimize AI for edge devices, all powered by the same technology stack used in Google’s own apps.

Hand Picked Video

In this video, we'll look at Elevenlabs Conversational AI Agents.

Top AI Products from this week

• Wondercraft - Create conversational podcasts using AI without recording. Wondercraft allows you to create engaging podcasts in minutes, with the use of AI voices, including your own. Create, translate, and share podcasts at scale - no recording or editing skills required.

• ALTAR - ALTAR is an AI-powered workspace that organizes everything you save - notes, images, links, and more - into a smart knowledge base. It learns what matters to you and helps you find answers, analyze insights, and create new ideas in just moments, helping you turn inspiration into action without the friction.

• Wispr Flow for iOS - Flow for iPhone is fast, natural voice dictation that works wherever you type—from iMessage to Gmail, Slack to Substack. Whether you’re on the go, juggling tasks, or just tired of typing with your thumbs, Flow helps you get stuff done 5x faster.

• Tilda Publishing - Create a website, landing page or online store for free with the help of Tilda modules and publish it on the same day. No knowledge of coding required.

• Promptmonitor - The most realistic text to speech and voice cloning software. The most compelling, rich, and lifelike voices for creators and publishers seeking the ultimate tools for storytelling.

• Browser.dating - The dating website matching people based on their browser history - by Dries Depoorter. The matching process uses AI.

This week in AI

• Meta AI Risks on Social Platforms - Meta's new AI tools on Facebook and Instagram raise concerns about privacy, misinformation, and user data security, prompting experts to call for stricter safeguards.

• Apple AI Lacks True Reasoning - Apple research finds large AI models like GPT and LLaMA mimic patterns, not logic, making them fragile and unreliable for tasks needing real reasoning3

• Record Labels Seek AI Music Deals - Universal, Warner, and Sony are negotiating to license music to AI startups Udio and Suno, aiming for fees, equity, and lawsuit settlements over AI-generated music use.

• OpenAI to Announce O3-Pro Soon - Some customers have been notified about the upcoming launch of OpenAI's O3-Pro, expected to be revealed very soon.

• ChatGPT H1 2025 Super-Assistant Push - OpenAI aims to evolve ChatGPT into a personalized AI super-assistant, focusing on deep user understanding, broad skills, and strong competitive moats in H1 2025.