Your curated collection of saved posts and media
wrist cam changes everything https://t.co/I2G3l3LW0w
We discovered an emergent property of VLAs like ฯ0/ฯ0.5/ฯ0.6: as we scale up pre-training, the model learns to align human videos and robot data! This gives us a simple way to leverage human videos. Once ฯ0.5 knows how to control robots, it can naturally learn from human video.
wrist cam changes everything https://t.co/I2G3l3LW0w
๐ข New issue brief: Have Chinese AI models pulled ahead of their global counterparts? Our latest brief analyzes Chinaโs diverse open-weight model ecosystem and examines the policy implications of their widespread global diffusion. https://t.co/zkdBKf7N5F https://t.co/MNugnSzaFb

MiniMax (Hailuo) Video Team Has Open Sourced VTP (Visual Tokenizer Pre-training)! VTP is a scalable pre-training framework for visual tokenizers, built for next-gen generative models. It challenges the conventional belief in Latent Diffusion Models that scaling the stage-1 visual tokenizer (model size, compute, data) doesnโt improve stage-2 generation. By combining representation learning + compressionโreconstruction, we show the first scaling curve connecting visual tokenizers with Diffusion Transformers! ๐งAnd the best part: no extra compute for the generator, but better generations simply by scaling up tokenizer training for better representation space! ๐๐ Merry pre-Christmas to the community!
itโs nice that there are still people out there who remember what the internet is really for https://t.co/q4wcfy63yK
itโs nice that there are still people out there who remember what the internet is really for https://t.co/q4wcfy63yK
https://t.co/nJEVS0p9wP

judge moved us one step closer to California suspending Tesla sales on a day Musk spent generating AI titty slop https://t.co/73f8g0Qf3W

Lightcone Infrastructure's 2026 fundraiser is live! We build beautiful things for truth-seeking and world-saving. We run LessWrong, Lighthaven, Inkhaven, designed AI-2027, and so many more things. All for the price of less than one OpenAI staff engineer ($2M/yr). More in ๐งต. https://t.co/Dqp9HxvoaM
@grinich @WorkOS ๐คฏ https://t.co/0SoDQefjcZ
Official rule for all AI labs: no more demoing your product with either telling the AI to โbook a trip for meโ or creating AI photos/videos of your companyโs CEO in crazy situations. Sorry, those are the rules now. https://t.co/PSnPM2P3u9
We just added @grok as a model in our LuxAlgo AI and chat history. More agentic strategies, better backtesting, deeper automation, and more are coming soon. We're still building. https://t.co/iYa4PK1IpH
NEW: German, Danish and Norwegian Medical Models, trained on 2 billion words of clinical data. ๐ European healthcare needs voice AI that works with non-native speakers, overlapping dialogue, regional accents, and medical terminology spoken under pressure. Generic speech recognition can't handle it. Ours can. ๐
Apple Music is coming to ChatGPT, OpenAI announces https://t.co/Huxq7X1d5b by @mvcmendes
Here are my lists. Unfortunately Apple doesnโt let me share most of them: https://t.co/PMgOlNfd2Z
Meta just released sam-audio https://t.co/ThdN1ZcCgB
Introducing SAM Audio: the first unified AI model that allows you to isolate and edit sound from complex audio mixtures. This could mean isolating the guitar in a video of your band, filtering out traffic noises, or removing the sound of a dog barking in your podcast, all with text, visual, and time span prompts.ย https://t.co/RiuUACGv80
Today, we're releasing BU-30B-A3B-Preview (OSS)๐ง For the first time, you can run hundreds of browser tasks reliably on $1 of compute๐ > 30B parameters > 3B active > SoTA quality at real-time speed Welcome to the future of web agents. Try it locally or on our cloud todayโ๏ธ HF link โ
We open sourced BU-30B-A3B-Preview. The best model for web agents ๐ง Here is how it performs compared to other models ๐ https://t.co/2oAc8kY5pO
Today, we're releasing BU-30B-A3B-Preview (OSS)๐ง For the first time, you can run hundreds of browser tasks reliably on $1 of compute๐ > 30B parameters > 3B active > SoTA quality at real-time speed Welcome to the future of web agents. Try it locally or on our cloud todayโ๏ธ HF
wow, this looks like a very solid open model by Xiaomi, competing with K2/DSV3.2 on benchmarks with fewer parameters. it's MIT licensed, with a very good tech report and base/thinking versions available it's using the same sliding window attention arch as gpt-oss (sink with SWA size = 128) but with much fewer global attention layers, multi-token prediction for speculative decoding support, and a new post-training distillation method. really seems like a beast at inference with day 0 @sgl_project support! really exciting
โก Faster than Fast. Designed for Agentic AI. Introducing Xiaomi MiMo-V2-Flash โ our new open-source MoE model: 309B total params, 15B active. Blazing speed meets frontier performance. ๐ฅ Highlights: ๐๏ธ Hybrid Attention: 5:1 interleaved 128-window SWA + Global | 256K context ๐ Pe
Microsoft just released TRELLIS 2 https://t.co/5IEzsmEfa8
ใใธใงใใใคใใโฆ๏ผ ่ชฐใงใ็ฐกๅใซใขใใคใซใชใผใใผใขใใช๐ใไฝใใกใใๆไปฃใๆฅใใ Manusใไฝฟใใฐใใณใผใใฃใณใฐไธๅใชใ ใชใผใใผใฎ่ฟฝๅ ใใซในใฟใใคใบใๆ้่จ็ฎใ๐ฐ ็ฅ่ญใผใญใงใใ็ดๆ็ใชๆไฝใ ใใงๆฌๆ ผ็ใชใขใใชใๅฎๆใใใชใใฆไฟกใใใใ๏ผ https://t.co/m8TtIP3c5Y
https://t.co/yOluGJEzI2
https://t.co/yOluGJEzI2
In the fast-changing world of AI, you're still in control. The latest @code release introduces new security and governance features, like the Private Marketplace, enterprise policy updates, tool auto-approval management, and more. ๐งต https://t.co/F8TzfMB72L
๐ Control extension access with Private Marketplace. Organizations can now curate which extensions are available to developers. Restrict installations to approved extensions only, ensuring compliance and security across your teams. https://t.co/PhG9mEmoS1 https://t.co/45TbjEur24
๐ Fine-grained URL approval gives you more control over external requests. When the fetch tool contacts a URL, you now see a two-step approval flow: 1๏ธโฃ Approve the domain to prevent sensitive data from reaching untrusted sites 2๏ธโฃ Review fetched content before it enters chat or follow-up tool calls
โ ๏ธ When enterprise policy disables agent mode, the Agents picker now clearly explains why. No more confusion for developers. They see exactly what policy restrictions apply. https://t.co/afENZ4KlJr
๐ Read the full release notes: https://t.co/wpQdDKcK9L Happy coding! ๐
we're open-sourcing a new frontier science eval in biology, chemistry, and physics. there are 2 tracks: olympiad level and advanced research level. as models become saturated on GPQA, this is a nice unsaturated alternative with clean test-time compute scaling. kudos to @MilesKWang for driving this release!
Training on tough benchmarks leads to better results on other benchmarks as well. Lots more to investigate here but this is the exact reason we work on these types of benchmarks. From the @XiaomiMiMo paper released today. https://t.co/nCq5GDJXyF
BREAKING: Databricks is raising over $4 billion in Series L funding at a $134 billion valuation. https://t.co/Qmx1iPZap4