'air head' is one of the first short films made using #Sora by OpenAI. the response so far has left us floating.🎈

Latest workaround for getting ChatGPT to spit out copyright protected imagery? Simply knowing another language. What a joke.

Something that drives me to distraction in discussion of AI alignment: someone will say "Oh, it's crucial we build systems with properties X, Y, Z to ensure safety". And different people have slightly different formulations of what X, Y, and Z ought to be, and argue over it

I struggle to remember most historical dates, even very approximately. I’m a spatial thinker so I’ve tried making a number of different visualizations of history to help me, and this is my latest and favorite: mapping many different time scales to my hand

Schedule-Free Learning github.com/facebookresear… We have now open sourced the algorithm behind my series of mysterious plots. Each plot was either Schedule-free SGD or Adam, no other tricks!

Two new AI releases by Apple today: 🧚‍♀️ OpenELM, a set of small (270M-3B) efficient language models. Weights on the Hub: Pretrained: huggingface.co/collections/ap… Instruct: huggingface.co/collections/ap… 👷‍♀️ CoreNet, a training library used to train OpenELM: github.com/apple/corenet

Introducing 🤗 Diarizers: a library for fine-tuning speaker diarization models 🗣️ Improve multilingual diarization performance by 30% with just 10 minutes of GPU compute time! ⚡️ The first release comes with training scripts, datasets and a Google Colab 🚀 Check it out! ⚒️

From Claude100K to Gemini10M, we are in the era of long context language models. Why and how a language model can utilize information at any input locations within long context? We discover retrieval heads, a special type of attention head responsible for long-context factuality

Data visualization when done well is so interesting

Excited to share that we're looking to hire a Senior Full Stack Eng at Ozu. We're working on cutting edge problems in storytelling with a very passionate and capable team. Take a look below for more details if you'd like to join us!

We've released a new library, fastdata, for high quality synthetic data generation! 🚀 Check out this deep dive thread and blog post from its creator explaining everything you need to know to get started:

1/ We're unwrapping the new architecture + benchmarks behind Playground v3 - our new foundation model focused on graphic design. This is our first step towards making a powerful AI graphic designer. It's state-of-the-art at text rendering, prompt understanding, and color

Have you ever wondered how to train an autoregressive generative transformer on text and raw pixels, without a pretrained visual tokenizer (e.g. VQ-VAE)? We have been pondering this during summer and developed a new model: JetFormer 🌊🤖 arxiv.org/abs/2411.19722 A thread 👇 1/

HOLY SHITT, Microsoft dropped an open-source Multimodal (supports Audio, Vision and Text) Phi 4 - MIT licensed! 🔥 > Beats Gemini 2.0 Flash, GPT4o, Whisper, SeamlessM4T v2 > Models on Hugging Face hub, integrated with/ Transformers! Phi-4-Multimodal: > Modalities: Integrates

Siglip needs registers For comparison, here's DINO-v2 with registers. It has five extra tokens for the model to work with: one CLS token and four "registers". Look at how smooth those attention maps are! No artifacts.

Vision models have been smaller than language models; what if we scale them up? Introducing Web-SSL: A family of billion-scale SSL vision models (up to 7B parameters) trained on billions of images without language supervision, using VQA to evaluate the learned representation.

📣Introducing the new OZU.ai - a new way to search for moments and scenes in film and tv. 🖤 Find themes you love, and moments that are in your head 🩶 Discover new shows, films, directors and actors ❤️ Change how you feel

Qwen is continuing their habit of state-of-the-art releases with 3 extraordinarily strong embedding models and 3 powerful reranker models, focusing on multilingual text retrieval and more. Details in 🧵

+1 for "context engineering" over "prompt engineering". People associate prompts with short task descriptions you'd give an LLM in your day-to-day use. When in every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window

GenAI isn't just a technology; it's an informational pollutant—a pervasive cognitive smog that touches and corrupts every aspect of the Internet. It's not just a productivity tool; it's a kind of digital acid rain, silently eroding the value of all information. Every image is no