LMStudio are using the upstream ggml implementation which is significantly better and well optimized. Looking at ollama's modifications in ggml, they have too much branching in their MXFP4 kernels and the attention sinks implementation is really inefficient. Along with other

There's a new tiny TTS model in town: Kitten TTS! 🐱 With just 15M parameters (<25 MB), it delivers impressive quality for its size, and can even run in real time without a GPU. So, I created a web demo for it: featuring text normalization, chunking, and real-time playback. 🤗

Realtime interactive generative models FTW! Announcing a new 🌊 of details and features for Magenta RealTime, the open weights live music AI model from GDM! * Live Jamming with audio input 🎤🎸🎵 * Personalize your own models 🔧 * Tech report 📜 Links below in the 🧵...

🚀 GLiNER x SmolLM: a new joint encoder-decoder architecture 🚀 We are excited to release a new kind of GLiNER model built with the mantra "you do the same things only once." Built on top of DeBERTa + Hugging Face SmolLM2 — full details below 👇

National security reframe (in order to get funding to solve the problem) - could an adversary be performing a death by 1000 spam calls attack to agitate and distract an entire population (of engineers)? 😅

LongSplat Robust Unposed 3D Gaussian Splatting for Casual Long Videos

Made a walkthrough vid for Magenta RealTime “Audio Injection”! The notebook takes ~10m to spin up, but totally worth it for the surreal experience 🎤💻🎧⁉️

Check out what you can do when you mix Gemini's world knowledge with the ability to show things visually. Multimodal communication abilities unlock new use cases!

Visual Story-Writing. While you write, our word processor visualizes the timeline, world map, and character relationships. Editing these visuals updates the story (e.g. drag a character on the map to move them). This summarizes our #UIST2025 paper. #HCI #LLMs #AI Thread 🧵 (1/8)

Introducing Alterego: the world’s first near-telepathic wearable that enables silent communication at the speed of thought. Alterego makes AI an extension of the human mind. We’ve made several breakthroughs since our work started at MIT. We’re announcing those today.

Tiny SOTA model release today: v3 of the Smart Turn semantic VAD model. Smart Turn is a native audio, open source, open data, open training code model for detecting whether a human has stopped speaking and expects a voice agent to respond. The model now runs in <60ms on most

Can we use video diffusion to generate 3D scenes? 𝐖𝐨𝐫𝐥𝐝𝐄𝐱𝐩𝐥𝐨𝐫𝐞𝐫 (#SIGGRAPHAsia25) creates fully-navigable scenes via autoregressive video generation. Text input -> 3DGS scene output & interactive rendering! 🌍mschneider456.github.io/world-explorer/ 📽️youtu.be/N6NJsNyiv6I

One more paper from the lab this week! 🥴 Multi-objective optimization of biological sequences isn’t limited to discrete diffusion. We present AReUReDi, our new framework that extends rectified discrete flows to provably converge to the Pareto front! Hope you're ready! 👇 📜:

#A quantum radio antenna leveraging Rydberg atoms achieves highly sensitive, all-optical detection of radio signals, enabling non-invasive, precise measurement without metal components. Nature Communications doi.org/g965h6 phys.org/news/2025-10-q…

Sakana AI has just leveraged their evolutionary code optimization system, ShinkaEvolve, to earn the 1st prize at ICFP Programming Contest 2025 🏆 ShinkaEvolve enabled up to a 10x speedup by evolving clever SAT encodings, unlocking solutions to far larger and more complex problems than

Today's a good day to brush up on some strategies for information dispersal across *multiple* cloud providers

Can we tell the human pose from an object alone? And the object given the human? With TriDi we can sample from any conditionl distribution of human, object and interaction. With the same model! Stop by our poster if you are at ICCV25! github.com/ptrvilya/tridi

Sweet signals for myelin: glucose sensing redirected to regeneration go.nature.com/3KUvnwf

🔎Did someone steal your language model? We can tell you, as long as you shuffled your training data🔀. All we need is some text from their model! Concretely, suppose Alice trains an open-weight model and Bob uses it to produce text. Can Alice prove Bob used her model?🚨