Charly Wargnier

🚨 @NVIDIA just quietly dropped an incredibly impressive speech recognition model that completely changes the math for local voice pipelines. Nemotron-3.5-ASR is a 0.6B parameter, open-source model built specifically for real-time streaming. What makes it so good: → 40+ supported languages → Cache-aware architecture (eliminates redundant audio computation) → Configurable latency (down to 80ms chunk sizes) → Emits beautifully punctuated, capitalized text automatically Because it’s so remarkably lightweight, it doesn't force you into a massive H100 dependency. It scales brilliantly on CPUs or widely available L40S GPUs. At its lowest latency setting, it handles ~17x more concurrent streams than previous 1.1B buffered models. HUGE win for devs building agent pipelines: you now have local, offline speech processing that is lighter, noticeably faster, and keeps data safely within your own security boundaries. 100% FREE and open-source. Repo and weights in 🧵↓

🚨 @NVIDIA just quietly dropped an incredibly impressive speech recognition model that completely changes the math for local voice pipelines. Nemotron-3.5-ASR is a 0.6B parameter, open-source model built specifically for real-time streaming. What makes it so good: → 40+ supported languages → Cache-aware architecture (eliminates redundant audio computation) → Configurable latency (down to 80ms chunk sizes) → Emits beautifully punctuated, capitalized text automatically Because it’s so remarkably lightweight, it doesn't force you into a massive H100 dependency. It scales brilliantly on CPUs or widely available L40S GPUs. At its lowest latency setting, it handles ~17x more concurrent streams than previous 1.1B buffered models. HUGE win for devs building agent pipelines: you now have local, offline speech processing that is lighter, noticeably faster, and keeps data safely within your own security boundaries. 100% FREE and open-source. Repo and weights in 🧵↓

Vibe-coding is the new gaming. The dopamine hit you get from actually creating something from scratch is unmatched.

an honest take from Nietzsche

@Cwjk1980 @US_Stormwatch BS

One of the worst heatwaves in European history is underway. Peak high temperatures forecast this week: France: 45°C / 113°F Monday-Tuesday London: 39°C / 102°F Amsterdam: 34°C / 93°F Berlin: 38°C / 100°F Paris: 41°C / 106°F

I’m organizing a run exclusively for climate skeptics tomorrow at 2 pm 😏

🚨 Anthropic’s Research Lead just dropped a masterclass on AI agents. "99% of our engineers run swarms of 300+ self-improving agents". The key? Closing the loop so models can verify their own work. In 20 minutes, they unpack continuous Claude loops, plan modes, and dynamic workflows. Skip the $300 courses. This is free, and better. Then read this ace Loop Engineering article by Rari 👇

@0xMortyx 🫡

@DataChaz That’s real gold, bookmarked

BAIDU JUST DROPPED AN ABSOLUTE GAME-CHANGER FOR DOCUMENT AI It’s called `Unlimited-OCR`, and it can literally transcribe an entire book in a single pass 🤯 Most vision models read a single page, forget the context, and eventually hit a wall where performance degrades and inference slows down. @Baidu_Inc built this on top of `DeepSeek OCR` but fixed the memory bloat with a single change to the attention mechanism. The design mimics how a human hand-copies a book. Instead of trying to hold the entire book in active memory, each token only looks at the current page plus the last 128 words. This creates a sliding window that keeps memory usage completely flat, no matter how long the output gets. The architectural shift delivers three massive upgrades for document parsing: → A fixed memory footprint → Steady generation speed on massive documents → The ability to process dozens of pages per pass The numbers back it up. Unlimited-OCR scores 93% on standard parsing benchmarks, beating the older baseline by a full six points. Even when pushed past 40 pages, the error rate stays under 0.11. More importantly, it maintains a flat speed curve where older models suffered a 35% slowdown. Free and open-source. Repo, weights and paper in 🧵↓

ARCHITECTS ARE GOING TO HATE THIS. Someone vibe-coded a full 3D building editor that runs entirely in the browser. No $50k/year Revit licenses. No desktop installations. It’s called Pascal Editor, and it just dropped a massive update adding full color mode for painting walls, furniture, and entire homes. The underlying tech is seriously impressive: → Built on React Three Fiber and WebGPU → Renders on your GPU at near-native speed → ECS-style architecture for high performance → Zustand state management with full undo/redo → Next.js frontend deployed simply as a web app You can stack levels, explode views, or solo individual floors. You can select zones, drag walls, and reshape slabs—all natively in 3D. The crazy part? It’s completely free and open source under an MIT license, already sitting at 16.9K stars. Tools that used to take years to build and cost thousands to license are now being shipped by independent devs using AI. Repo link in the 🧵↓

@notjazii fair! 😅

@DataChaz i'd say yes until you have to fix issues

5 👏 CENTS 👏 PER 👏 SECOND 👏 That's what @TopviewAIhq is now charging for Seedance 2.0 Mini in Credit Mode with the Ultra Annual pack 🤯 Or, if you’re not in a hurry, you can also generate credit-free for 365 days. The cost of shipping AI video has never been lower.

@asaukas sure. There's nothing wrong with copying this principle and improving upon it.

@DataChaz innovation badge goes to openclaw.

never ever leaving this app 🍿🍿🍿😎 The beef between Hermes and OpenClaw escalated quickly

@asaukas and your point is?

@DataChaz still a copycat

@Baidu_Inc Thx, appreciate it. and thanks for making this open source 🔥🔥🔥

@DataChaz Thanks Charly, this is a fantastic summary!

source: reddit.com/r/ClaudeAI/com…

♻️ If this resonated, share it. Someone may need this today! Follow me → @datachaz for daily drops on LLMs, agents, and data workflows

Paper → arxiv.org/abs/2606.23050