Inception Releases First Diffusion LLM

Hey there 👋

We hope you're excited to discover what's new and trending in AI, ML, and data science.

Here is your 5-minute pulse...

print("News & Trends")

Microsoft Releases Dragon Copilot, a Healthcare AI Assistant (7 min. read)

Microsoft introduces Dragon Copilot, a voice-driven AI assistant that integrates Nuance’s Dragon Medical One for real-time speech-to-text and DAX Copilot for ambient listening. It automates clinical documentation, retrieves patient data, generates structured SOAP notes, and executes EHR commands, streamlining workflows and reducing physician burden

LangChain Introduces LangGraph Swarm, a Python Library for Building Multi-Agent Systems (7 min. read)

LangGraph Swarm is a Python library for creating swarm-style multi-agent systems using LangGraph. In this architecture, agents dynamically hand off control based on their specializations, with the system remembering the last active agent to ensure seamless conversation continuity. Features include multi-agent collaboration and customizable handoff tools, facilitating communication between specialized agents.

Inception Unveils The First Diffusion-Based Large Language Models (5 min. read)

Inception, co-founded by Stanford professor Stefano Ermon, has emerged from stealth with a novel diffusion-based AI model. This architecture enables parallel text generation, offering up to 10 times faster performance and reduced computing costs compared to traditional large language models.

Sesame, a New AI Voice to Cross the “Uncanny Valley” (10 min. read)

Oculus co-founder Brendan Iribe’s new startup Sesame just launched a demo of its voice tech aiming to cross the "uncanny valley" of AI speech, showcasing a model that responds with genuine emotions and natural speech patterns. It’s Conversational Speech Model (CSM) incorporates contextual awareness, using conversation history to modulate tone, cadence, and emotional depth dynamically. Unlike traditional TTS systems, it goes beyond static prosody adjustments, achieving more natural, engaging interactions by adapting to user intent in real-time.

print("Applications & Insights")

BentoML: MLOps for Beginners (8 min. read)
BentoML streamlines MLOps, making model deployment fast and efficient. This beginner-friendly guide covers its core features, from packaging models to serving them with minimal effort. With built-in optimizations and seamless integrations, BentoML simplifies scaling and deployment, empowering data scientists to turn models into production-ready applications with ease.

Build App with Windsurf’s AI Coding Agents (Course)
Learn how to build real-world apps with AI coding agents using Windsurf in these short courses from DeepLearning.AI, featuring debugging, code generation, and best practices for project integration. Gain insights into the challenges of AI search and discovery, then apply these concepts using Windsurf to debug JavaScript, update legacy codebases, and building a Wikipedia analysis app that retrieves, processes, and visualizes data while learning to manage unexpected AI behavior.

Build a multi-agent flight booking crew using DeepSeek-R1 (15 min. read)
Leverage a multi-agent approach for flight booking using DeepSeek R1 to orchestrate specialized AI agents for tasks like searching flights, handling scheduling, and automating booking steps. This setup demonstrates how agent collaboration can reduce manual work, increase efficiency, and handle common pitfalls in flight booking workflows.

How to Test if Your Model’s Probabilities Are Good (Enough) (9 min. read)
Learn to evaluate the accuracy of your model’s predicted probabilities by examining calibration, employing metrics (e.g., Brier score), and visualizing reliability with calibration curves to ensure your model outputs meaningful confidence levels.

print("Tools & Resources")

TRENDING MODELS

Text-to-Video
Wan-AI/Wan2.1-T2V-14B
⇧ 151K Downloads
Wan2.1-T2V-14B is a 14-billion-parameter model that generates high-quality videos from textual descriptions, enabling users to create visual content based on text prompts.

Automatic Speech Recognition
microsoft/Phi-4-multimodal-instruct
⇧ 23.6K Downloads
Phi-4-multimodal-instruct is an advanced ASR model developed by Microsoft, designed to transcribe spoken language into text accurately, supporting various languages and dialects.​

Image-Text-to-Text
allenai/olmOCR-7B-0225-preview
⇧ 45.8K Downloads
olmOCR-7B-0225-preview is a model developed by AllenAI that converts images containing text into editable text formats, facilitating tasks like digitizing printed documents.​

Text Generation
perplexity-ai/r1-1776
⇧ 34K Downloads
r1-1776 is a text generation model by Perplexity AI, designed to generate human-like text for various applications, including chatbots and automated writing assistants.

TRENDING AI TOOLS

• 🔊 Speak: AI transcription and analysis software for audio, video, and text.

• 🤖 Copilot - Microsoft’s AI assistant, now available via MacOS

• 💬 ARI: Your AI search assistant for faster, deeper insights.

• ⚙️ Eppo: A modern experimentation platform for high-growth product teams.

That’s it for today!

Before you go we’d love to know what you thought of today's newsletter to help us improve the pulse experience for you.

See you soon,

Andres