Tom Sander @ICLR
@RednasTom
PhD student @AIatMeta (Meta FAIR) and @Polytechnique! ex: @École polytechnique (X18) , MVA. Interested in Private ML and watermarking
OpenAI may secretly know that you trained on GPT outputs! In our work "Watermarking Makes Language Models Radioactive", we show that training on watermarked text can be easily spotted ☢️ Paper: arxiv.org/abs/2402.14904 @pierrefdz @AIatMeta @Polytechnique @Inria
There's a lot of work now on LLM watermarking. But can we extend this to transformers trained for autoregressive image generation? Yes, but it's not straightforward 🧵(1/10)
Exciting PhD position open at FAIR in Paris. We are looking for a candidate to join our team and contribute to advancing the field of AI, especially reinforcement learning. Find more details and apply below. Feel free to reach out to me by email. metacareers.com/jobs/192266079…
Our first LLM coding agent: Devstral. It's the SoTA OSS model on SWE-Bench Verified by a large margin, while being only 24B. It's still a research preview specific to SWE tasks. For regular code completion Codestral might still be the best bet. Let us know your feedback!
Meet Devstral, our SOTA open model designed specifically for coding agents and developed with @allhands_ai mistral.ai/news/devstral
This is my first paper done at FAIR. We show that adaptive visual token segmentation, especially in subobject-level (i.e., subwords in images), enables VLMs to have a better and faster learning of image understanding! arxiv.org/pdf/2402.14327
Presenting at 3PM, poster #496. We just passed the 1k stars on GitHub🌟 long live neat open sourcing @pierrefdz
🎉Exciting news from @AIatMeta FAIR! We've released a Watermark Anything Model under the MIT license! It was announced yesterday: ai.meta.com/blog/meta-fair… Great project with @pierrefdz et al. ! We're close to hitting 1,000 stars on GitHub. Give it a try: github.com/facebookresear… 🚀
I am at ICLR this week, presenting Watermark Anything on Saturday, and another work on radioactive benchmarks at the WM workshop on Sunday. Reach out if you want to discuss Watermarking, Distillation Protecion, Memorization, or else :)
🎉Exciting news from @AIatMeta FAIR! We've released a Watermark Anything Model under the MIT license! It was announced yesterday: ai.meta.com/blog/meta-fair… Great project with @pierrefdz et al. ! We're close to hitting 1,000 stars on GitHub. Give it a try: github.com/facebookresear… 🚀
1/ Happy to share my first accepted paper as a PhD student at @Meta and @ENS_ULM which I will present at @iclr_conf: 📚 Our work proposes difFOCI, a novel rank-based objective for ✨better feature learning✨ In collab with David Lopez-Paz, @GiulioBiroli and @leventsagun!
And Watermark Anything got accepted at ICLR 2025 🎉
🎉Exciting news from @AIatMeta FAIR! We've released a Watermark Anything Model under the MIT license! It was announced yesterday: ai.meta.com/blog/meta-fair… Great project with @pierrefdz et al. ! We're close to hitting 1,000 stars on GitHub. Give it a try: github.com/facebookresear… 🚀
Makes you wonder what kind of evidence they have, and got me thinking about this Neurips paper: arxiv.org/pdf/2402.14904. Any chance it's radioactive watermarking @RednasTom @pierrefdz?
Distillation is a common practice in the AI industry but the concern was that DeepSeek may be doing it to build its own rival model, which is a breach of OpenAI’s terms of service. Here’s why: ft.trib.al/NydPc2b
Watermarking images treated as a segmentation task for much improved robustness 🧵📖 Read of the day, season 3, day 4: « Watermark Anything with Localized Messages », by @RednasTom, @pierrefdz et al from @AIatMeta As watermarking becomes more and more of a necessity to conform…
Presenting our spotlight poster Watermark Makes Language Models Radioactive at NeurIPS today from 4:30 - 7:30 pm poster session in East Hall #2411. Video, Poster and Slides: neurips.cc/virtual/2024/p… neurips.cc/media/neurips-… with @RednasTom
OpenAI may secretly know that you trained on GPT outputs! In our work "Watermarking Makes Language Models Radioactive", we show that training on watermarked text can be easily spotted ☢️ Paper: arxiv.org/abs/2402.14904 @pierrefdz @AIatMeta @Polytechnique @Inria
🎉Exciting news from @AIatMeta FAIR! We've released a Watermark Anything Model under the MIT license! It was announced yesterday: ai.meta.com/blog/meta-fair… Great project with @pierrefdz et al. ! We're close to hitting 1,000 stars on GitHub. Give it a try: github.com/facebookresear… 🚀
🔒Image watermarking is promising for digital content protection. But images often undergo many modifications—spliced or altered by AI. Today at @AIatMeta, we released Watermark Anything that answers not only "where does the image come from," but "what part comes from where." 🧵
I am in NeurIPS week :) Friday, Presenting our spotlight work: Watermarking Makes LLMs Radioactive ☢️ (arxiv.org/abs/2402.14904) Sunday, speaking at the image watermarking workshop about our latest Watermark Anything work (arxiv.org/abs/2411.07231) DM me if you’d like to chat :)
Check out our latest capabilities (web search, canvas, pixtral large, PDF upload) on chat.mistral.ai
Le Chat now includes image generation with FLUX1.1, web search, canvas, mistral large with vision capabilities, PDF upload, etc. And it's 100% free! chat.mistral.ai
Amazing work led by @RednasTom and @pierrefdz
☢️ Some news about radioactivity ☢️ - We got a Spotlight at Neurips! 🥳 and we will be in Vancouver with @pierrefdz to present! - We have just released our code for radioactivity detection at github.com/facebookresear….