post-training image/video models @meta
San Francisco
Joined July 2022
- Tweets10.6K
- Following918
- Followers997
- Likes42.8K
Mathieu Tuli retweeted
Art is very much a societal need. And the lack of empathy towards that nation is sad.
Mathieu Tuli retweeted
We gave our robots a sense of touch!
Excited to finally share this work with @AndrewEJXie, @leooeld, and the @rhoda_ai team!
Some technical details on the tactile sensor and robot policy: 🧵
Mathieu Tuli retweeted
Tomorrow is Canadian Thanksgiving. Or as we call it in Canada, Canadian Thanksgiving
Mathieu Tuli retweeted
companies are installing an insider threat, a trigger-able agent, inside their own systems.
we are building the infrastructure for us to be brought down from the inside.
we must onshore our intelligence.
we must secure our weights.
join @enclosure_inc
Mathieu Tuli retweeted
One of the worst forms of brainrot AI has cultivated is pessimism about basic research, ie the idea that important work can only happen inside a frontier/neo lab and only with 10k+ GPUs, so the rest should not even bother. What a bleak way to think about science. And it's false.
Mathieu Tuli retweeted
Yesterday, we released a cyberdefense strategy for superintelligence: Secure Acceleration.
Here are our calls to action:
1. We must build a competitive American open-source model and invent ways to detect AI sabotage
2. Containment should become a national security priority and we must develop capabilities to shut down untethered models
3. We must secure frontier model weights and RSI without impacting AI development
Secure the path to superintelligence.
secureacceleration.com
Mathieu Tuli retweeted
Right-Click Anything: a little demo combining Muse Spark, SAM 3.1 and Muse Image -- now all on @Meta Model API! Muse Spark captions the scene and generates noun-phrases. These are used to prompt SAM 3.1 which predicts instance masks for each object. Then, right-click any object mask as you would with text on any web page! “Look Up” hits Muse Spark web search for real product listings, “Copy” gives you the alpha-matted PNG, “Remove” paints the SAM 3.1 mask into the image and asks Muse Image to inpaint it away. SAM 3.1 also supports video so many more possibilities there too! Have fun combining models across Meta's API platform!
Mathieu Tuli retweeted
we're opening muse connectors to developers!
we have seen so much excitement in the developer community, integrating muse into everything from robots to mood lights and more.
plug your API into muse, then people can use your service just by asking for it. every request runs in a secure VM, and muse will ask before anything consequential.
come build with us! muse.ai/platform
Mathieu Tuli retweeted
📢 SAM 3.1 is now available on @Meta Model API - our official implementation with efficient inference built for the model. Detect, segment and track objects in images and video with noun phrase prompts, and easily combine with other Muse models with the same API surface, billing and auth. $2.50/1000 images, $0.20/1000 frames. Open weights still on Hugging Face!
Meta’s Segment Anything Model (SAM) 3.1 is now available on Meta Model API, giving developers a fast and lightweight model for detection, segmentation and tracking in a single call on inference tuned for SAM 3.1's architecture.
Use a short phrase to find objects in images and video. One API call returns detections, pixel-precise segmentation masks, and identity-preserving video tracks.
Learn more and start building: bit.ly/4rDArq9
Mathieu Tuli retweeted
Never trust Meta with your data.
Only trust <12 month old venture funded startups.
Mathieu Tuli retweeted
Don't care, build it
Toronto-Quebec high-speed rail project could cost more than $150-billion, internal document shows, @curryb reports theglobeandmail.com/politics…
Find out more at nationalnewswatch.com
Mathieu Tuli retweeted
The name Lake Ontario comes from the Wendat word Ontari’io, which, appropriately, means “the lake is beautiful, the lake is big”.
The name is more than 400 years old, predating both the Confederation of Canada and the Declaration of Independence of the United States of America.
We know that America is changing. Their trading relationships, their foreign policies, their national monuments, their hydronyms.
Canadians also know that naming reality means calling it Lake Ontario – then, now and always.
Mathieu Tuli retweeted
I mean literally just do anything. Anything else than scrolling twitter or papers :)
Mathieu Tuli retweeted
The final lecture of my course is an intro to character training! This is a topic that I've been quietly very invested in for ~18 months, as it:
* Has potential for high real world impact
* Clearly used extensively at frontier labs
* Almost no empirical literature exists
* More accessible on academic compute
This lecture covers what character training is, reviews model specs, constitutions, the differences, the motivations in real world events, some example research papers I like, and open questions in how it relates to post-training/model use generally.
Hopefully this brings more people into the field (and reach out if you have questions). It is one of the more research-y chapters in my book, but one that I felt needed the reference. There is still so little, educational content on the topic online.
0:00 Intro
6:22 Part 1: Fundamentals — character, constitutions, and model specs
19:21 Part 2: Character training in practice
23:23 Part 3: Character elicitation without gradient steps
28:03 Part 4: Open questions (and the end of the course)
32:27 The course, complete
Thanks for watching. No need to like and subscribe now that the course is done, you definitely wouldn't!
h/t to @_maiush for leading the technical work I got to do in the space, and @zafstojano for investing a lot of attention at this book chapter.
Mathieu Tuli retweeted
Introducing Muse Glimmer, an open-weight 30B-parameter model optimized for local, always-on agent workflows.
Muse Glimmer delivers strong performance on key agentic use cases and benchmarks compared with leading models in its size category, and is designed to run entirely on consumer hardware like a Mac or PCs with performant GPUs.
In keeping with our long tradition of sharing fundamental AI research, we’re releasing model weights under a permissive Apache 2.0 license.
🧵👇
