Yunan Jiang

📣 Blackwell sets a new inference speed world record — A single NVIDIA DGX B200 server with eight #NVIDIABlackwell GPUs can generate over 1,000 tokens per second (TPS) per user on the Llama 4 Maverick model, the largest and most powerful model in the AI at Meta Llama 4 family. ⚡⏱️ Additionally, a system with eight Blackwell GPUs can also deliver up to 72,000 tokens/second in a maximum throughput scenario. 🏆 Blackwell is the first platform to achieve this model performance, demonstrating how it delivers the best combination of throughput, output speed and accuracy for LLM token generation. 🏎️🏁 See our tech blog for details.

3

Podonos just raised $2.4 million in pre-seed funding, and the signal is loud and clear: the voice AI industry finally has someone measuring the sound, not just making the noise. Based in Los Gatos, Podonos is building the infrastructure layer that decides whether voice AI feels real, human, and ready for primetime. The industry is racing toward a projected $47.5 billion by 2034, but growth without standards is chaos. Podonos is making sure the hype actually holds up when you press play. At the center is ✦Soohyun Bae, PhD in Computer Engineering from Georgia Institute of Technology and Y Combinator W22 alumnus. His track record stretches from engineering roles at Google Maps to leading AR mapping at Niantic, plus co-founding Bobidi and TickTock AI. He has seen what happens when technology scales without proper #guardrails. With Podonos, he's betting the future of AI isn't just about how smart the models get, but how believable and trustworthy they sound. Podonos evaluates models across naturalness, similarity, emotion, recognition accuracy, pronunciation, tone, and resilience. Add in #personaconsistency, and suddenly you're not just testing machines, you're testing their ability to pass as human. The twist? They deliver results in under twelve hours, when the legacy approach drags on for two months. In AI time, that's the difference between relevance and irrelevance. Investors caught the beat. Serac Ventures, led by Kevin Moore, took the lead, backed by NAVER D2SF, the venture arm of South Korea's NAVER, and KAIST Venture Investment Holdings. Together, they've now backed Podonos twice, with this $2.4 million round stacking on top of a $750,000 raise in March 2025. Total funding sits at $3.15 million, and every dollar is pointed toward scaling #engineering, refining #analysistools, and launching into the U.S., Europe, and Southeast Asia. This isn't theory. Podonos already serves Resemble AI, Play AI, and Sanas AI, hitting six-digit ARR by November 2024 with just four employees. Their secret weapon is reach: 150,000 evaluators spread across nine languages and thirteen locales, forming a human-in-the-loop platform that blends speed, rigor, and global coverage. That reach lets Podonos deliver evaluations at a scale and pace that no competitor has matched. The strategy is ambitious: expand into high-demand sectors like #healthcare, #finance, #gaming, and #advertising, then move beyond voice into #multimodal model evaluation, spanning video AI and large language models. The mission isn't just to check if AI works, it's to decide if AI works for people. #Startups #StartupFunding #EarlyStage #VentureCapital #PreSeed #AI #AIInfrastructure #Voice #VoiceAI #VoiceTech #Infrastructure #Technology #Innovation #TechEcosystem #StartupEcosystem #Hiring #TechHiring If software engineering peace of mind is what you crave, Vention is your zen.

6

1 Comment

At IBC 2025, Monks shared how the shift from legacy workflows to real-time, AI-powered production is reshaping media. Built on the NVIDIA Media stack — including Holoscan for Media and Vision AI — this new approach unlocks faster insight, smarter storytelling, and scalable innovation across the industry. Explore how AI is redefining live content creation → https://bit.ly/3JmAepv

1

I don't post much but this is a big deal💥: Lightricks just open-sourced LTX-2 - the first production-ready audio-video generation model with truly open weights. ✦ Native audio + video ✦ Fine-grained control ✦ Runs locally on NVIDIA RTX GPUs. LTX-2 is the backbone of LTX Studio, so I've seen firsthand what it's capable of. Huge shoutout to the team behind this release - excited to see what the community builds with it.

47

1 Comment

We released a new VibeVoice model, VibeVoice ASR, a unified speech-to-text model designed to handle 60-minute long-form audio in a single pass, generating structured transcriptions containing Who (Speaker), When (Timestamps), and What (Content), with support for Customized Hotwords. It redefines ASR in the era of LLMs, and the new ASR will be a foundation to unlock the potential of new AI-Native hardware, memory systems for AI, and many other scenarios. Please have a try at https://lnkd.in/dNCCzQiN Key features: -         60-minute Single-Pass Processing: Unlike conventional ASR models that slice audio into short chunks (often losing global context), VibeVoice ASR accepts up to 60 minutes of continuous audio input within 64K token length. This ensures consistent speaker tracking and semantic coherence across the entire hour. -         Customized Hotwords: Users can provide customized hotwords (e.g., specific names, technical terms, or background info) to guide the recognition process, significantly improving accuracy on domain-specific content. -         Rich Transcription (Who, When, What): The model jointly performs ASR, diarization, and timestamping, producing a structured output that indicates who said what and when. We are continuously pushing the frontier of voice AI. More release on the way. Stay tuned! https://lnkd.in/dATPu9nw

380

8 Comments

Conditioning with 3d bounding boxes is an intuitive step for editable scene generation.

26

The next wave in AI is here: Real-time conversational intelligence! PersonaPlex-7B-v1 by NVIDIA represents a major shift from text-based AI to natural speech interactions. We're moving beyond chatbots to truly conversational AI agents. This trend signals the democratization of voice AI - from enterprise call centers to personal assistants, the barrier to entry just dropped significantly. Your thoughts on this shift toward voice-first AI experiences? 🎯 #VoiceAI #AITrends #ConversationalAI #TechInnovation

17

* Real-time, infinite video generation is here. Meet MotionStream. Most motion-controlled video models still do “render-and-wait.” MotionStream flips that: sub-second latency, up to 29 FPS on a single GPU, and interactive control (draw paths, drag objects, steer the camera) with constant cost per frame—so you can stream forever. Why it’s a big deal Teacher=>Student distillation: a strong bidirectional diffusion “teacher” gets distilled into a causal, streaming “student” via Self-Forcing + Distribution Matching Distillation. Translation: high quality, real-time speed. No more drift on long clips: clever sliding-window causal attention with attention sinks + rolling KV cache keeps quality stable on arbitrarily long videos. Precise yet natural motion: lightweight sinusoidal track embeddings + joint text+motion guidance for both trajectory fidelity and realistic dynamics. Numbers that pop ≤ ~0.4s latency, ~29 FPS @ 480p with an optimized tiny VAE; ~10 FPS @ 720p on one H100. SOTA motion following at speeds ~100× faster than prior motion-controlled pipelines. What you can do Live motion transfer from tracker outputs Drag-to-direct objects and characters Real-time camera control (single image => smooth 3D-like moves) - all as you create, not after. If you work in VFX, gaming, ads, or creative tooling, this turns video generation from a post-process into a live medium. 🚀 🔗 Paper: “MotionStream: Real-Time Video Generation with Interactive Motion Controls” (Nov 2025). #AI #VideoGeneration #DiffusionModels #VLM #ComputerVision #GenAI #Realtime #CreativeTech #Autoregressive #StreamingAI

8

Vector embeddings are growing at an alarming exponential rate and tomorrow's AI is going to rely on fast and efficient ingest. OpenSearch 3.0 now features vector ingest on the GPU, slashing indexing time by up over 9x and cost by nearly 4x! This enables fast day-0 indexing, iteration, and re-indexing to avoid model lock-in. What's great is that the indexes are deployed to the CPU for search.

9

Introducing Streaming Workflows for AI agents. Akka has released workflow notifications, that leverage Akka streams to publish sequences of notifications as real time data streams. Workflows allow you to have long running, fully controlled orchestrated processes. When workflow steps call long-running AI agents, those agents can stream their responses as chunks. With other systems, that real time response from an agent would have to be piped directly to an HTTP endpoint, bypassing the control and orchestration of durable execution. But with Akka, you can bridge your agent's token stream into the workflow's notification system. This allows clients to receive real-time LLM responses while still benefiting from workflow orchestration. All of Akka's map-reduce capabilities for the streamed behavior are still available. Now you can have real time streaming data, while still preserving the distributed trace and control of Workflows.

55

1 Comment