Sign in to view Julien’s full profile
or
Already on LinkedIn? Sign in
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Sign in to view Julien’s full profile
or
Already on LinkedIn? Sign in
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Zurich, Zurich, Switzerland
Sign in to view Julien’s full profile
Julien can introduce you to 10+ people at Microsoft
Join with email
or
Already on LinkedIn? Sign in
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
5K followers
500+ connections
Sign in to view Julien’s full profile
or
Already on LinkedIn? Sign in
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
View mutual connections with Julien
Julien can introduce you to 10+ people at Microsoft
Join with email
or
Already on LinkedIn? Sign in
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
View mutual connections with Julien
or
Already on LinkedIn? Sign in
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Sign in to view Julien’s full profile
or
Already on LinkedIn? Sign in
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
About
Welcome back
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
New to LinkedIn? Join now
Activity
5K followers
-
Julien Valentin shared thisToday progress on Photorealistic Avatars is largely quantified by metrics like PSNR, SSIM or LPIPS which are not well suited for evaluating the quality of communications when using Photorealitic Avatars. The CVPR 2025 Photorealistic Avatar Challenge provides a new evaluation methodology to address current limitations and will start tomorrow with an initial data drop! See Ross’ post below and register! Looking forward to seeing you at #CVPR2025!Julien Valentin shared thisMicrosoft, Meta, and TU Darmstadt are organizing a new CVPR workshop! The CVPR 2025 Photorealistic Avatar Challenge is intended to stimulate research in the field of photorealistic avatars. The challenge provides a test set and new methodology to subjectively evaluate photorealistic avatars for news anchor and telecommunication scenarios. Test subjects will be sitting or standing but only the upper half of the body is rendered. Speech, facial emotions, head turning, and hand gesture sequences are included. For more details please see: https://lnkd.in/gA2Dtm8V #CVPR2025Photorealistic Avatar Challenge CVPR 2025 - Microsoft ResearchPhotorealistic Avatar Challenge CVPR 2025 - Microsoft Research
-
Julien Valentin reposted thisJulien Valentin reposted this🧿🧿 "Look Ma, no markers" Capture 🧿🧿 👉#Microsoft unveils the first technique for marker-free, HQ reconstruction of COMPLETE human body, including eyes & tongue, without requiring any calibration, manual intervention or custom hardware. Impressive results! Repo for training & Dataset💙 𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬: ✅Novel SOTA holistic 3D human reconstruction ✅Body shape/pose + face shape/expression ✅Hand & tongue articulation + eye gaze ✅Suitable for mono & multi-view scenario ✅Data pipeline for generating synthetic data ✅Body, face and hands dataset released! #artificialintelligence #machinelearning #ml #AI #deeplearning #computervision #AIwithPapers #metaverse #LLM 👉Discussion https://lnkd.in/dMgakzWm 👉Paper arxiv.org/pdf/2410.11520 👉Project https://lnkd.in/duQ2xzQH 👉Repo https://lnkd.in/du4qtzRV
-
Julien Valentin reposted thisJulien Valentin reposted thisLook Ma, no markers: Holistic performance capture without the hassle Motion capture is usually a real pain - involving marker suits, loads of cameras, painstaking calibration, and hours of cleanup after that! Our method allows for capture of the body, face and hands in a single take and works in-the-wild without suits. You can use any cameras (including only one) and don't need to calibrate - and the whole method uses only synthetic training data! I'll be presenting the work at SIGGRAPH Asia in Tokyo in December, but you can already access the three datasets we're releasing alongside the publication. Find out more here: https://aka.ms/SynthMoCap Huge thanks to my co-authors: Fatemeh Saleh, Sadegh Aliakbarian, Lohit Petikam, shideh rezaeifar, Louis Florentin, shideh rezaeifar, Louis Florentin, Zafiirah Hosenie, PhD, Tom Cashman, Julien Valentin, Darren Cosker and Tadas Baltrusaitis
-
Julien Valentin reposted thisJulien Valentin reposted thisHuman reconstruction always limited to tight-fitting clothing 👕? Not anymore! In our #ECCV2024 paper ReLoo, we propose a method to reconstruct the detailed geometry and appearance of humans dressed in loose garments 💃 from only monocular in-the-wild videos! 🥳 Our key insight is first to build a layered neural human representation that separately models the shape of the inner body and outer clothing. On top of the layered representation, we introduce a virtual bone deformation module to accurately track the large surface dynamics. Our method ReLoo achieves consistent and high-quality reconstructions of clothed humans dressed in diverse loose outfits, even under challenging human poses. ReLoo is a joint work contributed by Tianjian Jiang, Manuel Kaufmann, Chengwei Zheng, Julien Valentin, Jie Song, and Otmar Hilliges! Project page: https://ait.ethz.ch/reloo Paper: https://lnkd.in/e6nNZYeu Code: https://lnkd.in/eDfbsrK8 (scheduled in Dec.) ReLoo will be presented at #ECCV2024 Poster Session 7 at 264. You're welcome to drop by and chat with us!
-
Julien Valentin reposted thisJulien Valentin reposted this🐳🐳MultiPly: in-the-wild Multi-Pax from Mono🐳🐳 👉ETH (+#Microsoft) announced MultiPly, a novel framework to reconstruct multiple people in 3D from monocular in-the-wild videos. It's the new SOTA over the publicly available datasets and in-the-wild videos. Source Code announced, coming💙 𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬: ✅Multiple detailed 3D humans from in-the-wild ✅Novel robust instance segmentation approach ✅Clean separation between interacting people ✅Accurate confidence-guided optimization ✅Temporal/spatial coherent 3D reconstructions #artificialintelligence #machinelearning #ml #AI #deeplearning #computervision #AIwithPapers #metaverse 👉Discussion https://lnkd.in/dMgakzWm 👉Paper https://lnkd.in/eSdFh_-i 👉Project https://lnkd.in/ebCeh7MJ 👉Repo https://lnkd.in/eGWeGMZF
-
Julien Valentin reposted thisJulien Valentin reposted this📢 📢 Are you working on - Human reconstruction./pose estimation - Object pose/reconstruction. - Interaction of humans and environments - Robotics control - Other topics related to humans, objects, scenes Then submit to our RHOBIN workshop @CVPR Deadline is March 20 (next Thursday)! https://lnkd.in/dCJjEsjq
-
Julien Valentin reposted thisJulien Valentin reposted thisThe Microsoft Mixed Reality and AI Zurich Lab is growing. We are looking for several Research Scientists and Software Engineers who are excited about bringing together Spatial Computing and AI. Apply to the links below if interested: https://lnkd.in/dJQmiwnf https://lnkd.in/dQCDaTzP
-
Julien Valentin shared thisIf you are interested about advances on efficient reconstruction and control of both static and dynamic scenes using point clouds, check out “Dynamic Point Fields: Towards Efficient and Scalable Dynamic Surface Representations”, which is at #ICCV2023 this week! Great collaboration with Sergey Prokudin, Qianli Ma, Maxime Raafat, and Siyu Tang! Project page: https://lnkd.in/ei_dVHe7 Video: https://lnkd.in/eHH5a6Ah Code: https://lnkd.in/ebHdXqbB Oral session: Thursday 5th 09:00 AM-10:30 AM in Room "Paris Sud" Poster session: Thursday 5th 10:30 AM-12:30 PM in Room "Nord" - 156
-
Julien Valentin shared thisJulien Valentin shared thisAre you interested in helping to build the Industrial Metaverse? Our MR&AI Zurich Lab is growing. We have several open positions: #Microsoft #ai #mr #industrialmetaverse #spatialai #zurich https://lnkd.in/eDRvzyDU https://lnkd.in/eYkaCTWA https://lnkd.in/eAb4TkNzResearch Scientist - Mixed Reality and AI in Zürich, Zürich, Switzerland | Research, Applied, & Data Sciences at MicrosoftResearch Scientist - Mixed Reality and AI in Zürich, Zürich, Switzerland | Research, Applied, & Data Sciences at Microsoft
-
Julien Valentin liked thisJulien Valentin liked thisEpstein files, if this is the stuff Trump has allowed us to see, I dread to think what bomb shells lie under the half he won't release. we're only seeing the tip of the ice berg. but justice must come, and hopefully will, to all those victims no matter how powerful the perpetrators
-
Julien Valentin liked thisJulien Valentin liked this#AIdubbing was highlighted on Meta’s earnings call last week as part of how AI is driving performance into 2026. It’s amazing to see our work evolving from research into a product benefiting hundreds of millions users at global scale! #MetaRealityLabs "... with hundreds of millions of people watching AI translated videos every day. This is already driving incremental time spent on Instagram ..." - Meta CFO Susan Li
-
Julien Valentin liked thisJulien Valentin liked thisGoogle DeepMind just released one of the most mind-blowing general-purpose world models I've ever seen: Genie 3. 🤯 "Given a text prompt, Genie 3 can generate dynamic worlds that you can navigate in real time at 24 frames per second, retaining consistency for a few minutes at a resolution of 720p." Looks like a quick way to make a mid-video game or VR world to explore BUT the implications are much more profound ... it essentially creates unlimited simulation training environments for AI agents, and I quote from Google directly, "this model is a key stepping stone to AGI." Video from some of my favorite creations last night. #google #agi #ai #worldmodels #virtualreality
-
Julien Valentin liked thisJulien Valentin liked thisWe release Action100M, the hero behind VL-JEPA. It is a large dataset with O(100 million) dense action annotations on HowTo100M procedural videos. We hope it serves as a robust data foundation to advance physical world modeling research. Action100M improved and scaled the pipeline introduced in our prior work Vision Language World Model (https://lnkd.in/em8uQZbK). The pipeline leverages a family of frontier foundation models to first generates hierarchical Tree-of-Captions (100+ nodes per video), then aggregates the information and extracts annotations via LLM Self-Refine. We show that scaling on Action100M with VL-JEPA (https://lnkd.in/gDaTBjkJ) yields consistent gains on zero-shot action recognition benchmarks (SSv2, EPIC-KITCHENS, EgoExo4D, COIN, etc.). It also enables strong zero-shot performance on video retrieval datasets (ActivityNet, YouCook2, etc.). See our paper on arXiv (https://lnkd.in/eAbtruuK) -- Action100M: A Large-scale Video Action Dataset by Delong Chen, Tejaswi Kasarla,Yejin Bang, Mustafa Shukor, Willy Chung, Lei Yu, Allen Bolourchi, Théo Moutakanni, and Pascale Fung Dataset can be acceesed from this repo: https://lnkd.in/eS2ZHyVS
-
Julien Valentin reacted on thisJulien Valentin reacted on thisI’ve decided to leave Meta. I’m grateful for the people, teams, and friendships I built there - they’re what made the experience meaningful. At the same time, I no longer felt aligned with the company culture or direction. Over time, the environment became increasingly chaotic, with a lack of clear, compelling vision, and a shift toward a more aggressive and fear-driven leadership style. As a result, I gradually lost a sense of purpose in what we were building, as values felt increasingly disconnected from everyday reality. I strongly believe culture is foundational to sustainable, high-performing organisations. When empathy, inclusion, psychological safety and ethical clarity are deprioritised, motivation and long-term engagement suffer, for people and for the business. I’m now taking a career break to slow down, reflect, and reconnect with life beyond work. I’m open to thoughtful conversations during this time, with the clear understanding that culture will be paramount in my next role - lived values, empathetic leadership, and genuine inclusion are non-negotiable for me. I’m particularly interested in co-founding startups or helping companies build and expand a presence in Zürich. Creating strong teams from the ground up and leading them through robust execution toward product excellence is one of my key strengths, and over the years I’ve developed a trusted network of outstanding engineers here. For my next chapter, I want to help build something meaningful with clear purpose, real impact, and a strong foundation in how it’s built. Grateful to everyone who made this chapter matter 💙
-
Julien Valentin liked thisJulien Valentin liked thisTurn-taking is one of the hardest unsolved problems in conversational AI. Today, we’re introducing Sparrow-1, a conversational-flow model built to solve it. Human conversation is driven by timing and nuance, similar to a dance. People subconsciously coordinate rhythm using tiny semantic and acoustic cues that signal when to speak, pause, or continue. Most conversational AI systems do not model this well. They wait for silence and rely on fixed thresholds, leading to delays, interruptions, and broken immersion. This has created a false tradeoff between speed and correctness. Sparrow-1 is built for how humans actually talk, not the tradeoffs machines have accepted. Instead of reacting after speech ends, Sparrow-1 continuously predicts floor ownership at the frame level, operating directly on streaming audio with a persistent state. In real-world evaluations, Sparrow-1 achieved: • 100% precision and recall • Zero interruptions • 55ms median response latency No tradeoff between speed and correctness. Sparrow-1 is now available for GA across all Tavus products, including our API, PALs, the Tavus platform, and the live demo at tavus.io. Read the full research blog here: https://lnkd.in/gvE7Z-FP
-
Julien Valentin liked thisJulien Valentin liked thisHonored to have been named a #IEEE #Fellow for "contributions to human motion analysis and deepfake technology for creative use and responsible safeguarding". Thank you all for your support and collaboration over the years. Excited to further our work with my amazing colleagues in 2026 at Google DeepMind and the wider ecosystem on these areas and more!
Experience & Education
-
Meta
****** ***** ********* *****
-
*********
********* ********* *****
-
******
****** ******** *********
-
********** ** ******
****** ** ********** ******* ******** ****** * ******* ******** undefined
-
-
******* ****** ** ******** *** **********
********* *** undefined
-
View Julien’s full experience
See their title, tenure and more.
Already on LinkedIn? Sign in
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Languages
-
French
Native or bilingual proficiency
-
English
Native or bilingual proficiency
Organizations
-
Microsoft Research
Research intern
-
View Julien’s full profile
-
See who you know in common
-
Get introduced
-
Contact Julien directly
Explore more posts
-
Julia Proskurnia
Google • 3K followers
Why "On-Device" is the most ethical AI strategy. We talk a lot about the size of models. Parameters. Trillions of tokens. The race for AGI. But in my academic life at EPFL and my work in Speech, I’ve always been obsessed with the opposite end of the spectrum: Accessibility. When we published our research in Nature Communications on AI for social projects, one conclusion was clear: If AI requires a massive data center and a high-speed fiber connection to run, it excludes half the world. Cloud models are undeniably impressive and they are pushing the boundaries of what is possible. But they shouldn't be the only way we deploy intelligence. I'm seeing a resurgence of interest in on-device solutions, not to replace the cloud, but to handle the sensitive, immediate, and personal tasks where the edge shines: - Privacy: Processing data right where it’s created. This is crucial for multimodal data (voice, video) that is highly identifiable and shouldn't leave the user's hand. - Latency: Real-time processing that feels like a conversation, not a transaction. - Energy Efficiency: Running a small, optimized model locally is often far more sustainable than round-tripping every query to a power-hungry data center. - Reliability: True accessibility means working offline. AI should be helpful even when the signal drops or the fiber hasn't been laid yet. Better locally-run alternatives make these powerful tools accessible to everyone, even for problems and data that are highly sensitive. I’m planning to dive deeper into this myself by experimenting with some of the latest open-weights models specifically for local deployment. If you’ve been testing small language models (SLMs) or edge-optimized architectures recently, I’d love to hear about your experience. What frameworks or models are you finding most promising for mobile/edge/small server hardware right now? #AIforGood #OnDeviceML #NLP #EthicalAI #TechForGood #EdgeAI #SustainableTech #OpenWeights
7
6 Comments -
Simron Patel
Chemetrian • 1K followers
🚀 The authors from Roche and EPFL claim their Minerva system can handle 96-reaction batches while existing tools like EDBO max out around 16 reactions leveraging Bayesian Optimization. They claim it is largely due to the difference in acquisition function choices. 🧠 The paper suggests EDBO's algorithm crashes when you try to scale up because it uses an acquisition function, q-EHVI, that has exponential computational complexity. Minerva switches to different algorithms (q-NEHVI, q-NParEgo) that supposedly scale much better. But I'm curious about the real-world impact? 🧪 I understand EDBO has its limitations but has a switch to these acquisition functions solved these problems? I would think there is also a need to scale the compute power of the server running these jobs. 🧠 The literature also claims that one should use DFT descriptors versus one-hot encoding to represent categorical variables. Now in theory this would build dense vectors typically better for all ML models versus OHE's sparsity issues but one should consider the trade-offs. The time and power to compute certain levels of theory for physics-based descriptors can be quite long and the uplift might be negligible versus risking a debatably less accurate objective function due to OHE. Abigail Doyle Would love to get the community's thoughts! Relevant link in the comments Chemetrian #Chemistry #MachineLearning #AI #DrugDiscovery #ProcessOptimziation #SmallMoleculeDiscovery
6
1 Comment -
Pragna Thotakura
Velsera • 1K followers
FunctionGemma, a new 270M parameter language model from Google, designed specifically for function calling and tool execution — not general chat. Why is FunctionGemma special? 🔹 Extremely lightweight FunctionGemma can run in ~550MB RAM, making it possible to deploy on laptops, edge devices, and even phones — all without large GPUs. 🔹 Purpose-built for function calling Unlike general LLMs, this model is trained for text-only tool calling, making it ideal for automation, agents, and API-driven workflows. 🔹 Easy local deployment The model is released in GGUF format, so it works seamlessly with llama.cpp and Ollama using quantized versions (Q4/Q5/Q8). 🔹 Fine-tuning ready from day one Thanks to Unsloth’s day-zero support, you can: Run FunctionGemma on CPU or GPU Fine-tune it locally using Colab notebooks Choose full fine-tuning or LoRA What can you fine-tune it for? FunctionGemma is meant to be customized for specific tasks, such as: Multi-turn function-calling agents Mobile actions (calendar events, reminders, flashlight, timers) Task-specific automation and workflows Lightweight, offline AI assistants This model clearly shows a shift toward small, focused, and efficient AI — where models are trained to do one thing well and can be adapted quickly to real-world use cases. Model (GGUF): https://lnkd.in/gxVuBhZA Docs & fine-tuning guides: https://lnkd.in/gchpAjUF Run and Deploy LLMs on your iOS or AndroidPhone: https://lnkd.in/grk72Bey
5
-
Dr. Stefan Bartels
Universitätsklinikum Hamburg… • 179 followers
🚀 New Publication Alert 📄 From Text to Data: Open-Source Large Language Models in Extracting Cancer-Related Medical Attributes from German Pathology Reports I’m excited to share our latest study, now published open access: 👉 https://lnkd.in/eYye9pnu In this work, we explored how open-source large language models (LLMs) can extract structured oncological information from unstructured German pathology reports — fully locally and in compliance with strict data privacy regulations. 🧠 We evaluated models like Llama 3.3 70B and Mistral Small 24B, combining retrieval-augmented generation (RAG) with tailored prompt engineering. Our results show that even smaller models can reach high accuracy (F1 > 0.90) with the right strategy — making real-world clinical use both feasible and resource-efficient. 🏥 The methodology is about to be implemented in our local clinical cancer registry at the #UKE to support automated tumor documentation and improve data quality. 🛠️ The full code and prompt templates are available open source: 👉 https://lnkd.in/e6qppChq This is a step toward practical, privacy-aware AI integration in oncology — built on reproducible tools and focused clinical needs. #MedicalAI #RAG #CancerRegistry #UCCHamburg #LLMs #OncologyInformatics #ClinicalNLP #DigitalHealth #OpenSourceAI #HealthData #Pathology #AIinHealthcare
13
3 Comments -
Luis Cisneros
Serelora • 3K followers
Yann LeCun on architectures that could lead to AGI LLMs can take us only so far. "If you are interested in human-level AI, don’t work on LLM Abandon generative models in favor joint-embedding architectures Abandon probabilistic model in favor of energy-based models Abandon contrastive methods in favor of regularized methods Abandon Reinforcement Learning in favor of model-predictive control Use RL only when planning doesn’t yield the predicted outcome, to adjust the world model or the critic."
1
4 Comments -
Antonio Mallia
Seltz • 4K followers
🚀 Reranking that's smarter, not just stronger Our paper "E2Rank: Efficient and Effective Layer-wise Reranking" was presented at ECIR 2025 — the result of a collaboration between Sapienza and Pinecone with Cesare Campagnano, Jack Pertschuk, and myself. 🔍 What is E2Rank? It's a cross-encoder-based reranker designed to adapt its computational budget layer by layer — starting light and only getting deeper for the most promising candidates. That means: 🔄 Fewer FLOPs, smarter compute reuse 🎯 SOTA effectiveness across TREC and BEIR ⚡️ Lower latency, even with large models like DeBERTa v3 📊 Instead of one-size-fits-all reranking, E2Rank introduces a multi-step layerwise strategy: progressively refine the candidate set using deeper transformer layers at each step — reusing hidden states, saving compute, and keeping latency low. 🏅 Bonus: it's on the Pareto frontier for efficiency vs. effectiveness. #InformationRetrieval #NeuralSearch #ML #RAG #Retrieval #ECIR2025 #Pinecone
28
1 Comment -
John Jorgensen
Philips • 169 followers
OpenAI's paper on Why Language Models Hallucinate is pretty interesting. Paper here: https://lnkd.in/eSb6itBJ The TLDR is that during fine tuning the reward function, similar to multiple choice exams values any guess, right or wrong, higher than "I don't know". Maybe hallucinations are not intrinsic to the architecture of LLM's. This seems obvious in retrospect and I imagine it will result in better models for a lot of use cases.
1
-
Philippe Bich
2K followers
𝗦𝗜𝗡𝗤 🇨🇭 𝗔𝗽𝗲𝗿𝘁𝘂𝘀: Open Source AI, Made in Switzerland. We’re proud to announce a 𝗻𝗲𝘄 𝗾𝘂𝗮𝗻𝘁𝗶𝘇𝗲𝗱 𝘃𝗲𝗿𝘀𝗶𝗼𝗻 𝗼𝗳 𝘁𝗵𝗲 𝗔𝗽𝗲𝗿𝘁𝘂𝘀-𝟴𝗕 model using our novel 𝗦𝗜𝗡𝗤 method! This release combines Apertus, an open-source large language model born from EPFL, Eidgenössische Technische Hochschule Zürich, and the Swiss National Supercomputing Centre (CSCS), with 𝗦𝗜𝗡𝗤, our Zurich-born quantization method released under the Apache 2.0 license. Thanks to the integration with fast kernels, you can now experience 𝗵𝗶𝗴𝗵-𝘀𝗽𝗲𝗲𝗱 𝗶𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲 using our 𝗵𝗶𝗴𝗵-𝗾𝘂𝗮𝗹𝗶𝘁𝘆 𝟰-𝗯𝗶𝘁 𝗺𝗼𝗱𝗲𝗹𝘀. I am personally excited to continue contributing to the Swiss open-source AI ecosystem and this is just the beginning. New quantized versions, including for the 70B model, are coming soon! 👉 New SINQ-quantized Apertus model: https://lnkd.in/d5hvFVDC 👉 SINQ GitHub Repo: https://lnkd.in/dctt5jK2 ⭐ Don’t forget to star our repo and join the journey toward fast, open, and collaborative AI made in Switzerland.
104
3 Comments -
Ken Ross
Microsoft • 2K followers
Are you adding AGENTS.md or CLAUDE.md to all your repositories because "everyone says so"? You might want to hold that thought. 🛑 A new paper from researchers at ETH Zurich and LogicStar.ai just dropped the first rigorous evaluation of repository-level context files, and the results are... counter-intuitive. While industry leaders are pushing these files to help AI agents navigate code, this study reveals that "more context" doesn't always mean "better performance". Using a novel benchmark called AGENTBENCH, they tested top-tier agents like Claude Code, Codex, and Qwen Code across real-world GitHub issues. The TLDR: Surprisingly, LLM-generated context files actually reduced task success rates in most settings while spiking inference costs by over 20%. While these files successfully prompt agents to follow specific instructions and test more thoroughly, they often make the core task harder by introducing unnecessary requirements and redundant information. The data suggests that for now, we should skip the automated /init commands and stick to minimal, human-written requirements. Is your "context engineering" actually hurting your agent's productivity? Read the full paper to see the detailed breakdown of the AGENTBENCH results and why our current recommendations might be leading agents astray. 🔗 https://lnkd.in/eDf-XWnY #AI #SoftwareEngineering #LLMs #CodingAgents #ContextEngineering #Research #Github
5
-
Dr. David J. Cox
Mosaic Pediatric Therapy • 5K followers
+1000 points for the Swiss!! "Switzerland launched Apertus, a national LLM developed by EPFL, ETH Zurich, and the Swiss National Supercomputing Centre as an alternative to ChatGPT, Meta’s Llama models, and DeepSeek. The model comes in two sizes — 8 billion and 70 billion parameters — and was trained on 15 trillion tokens across more than 1,000 languages. 40 percent of Apertus’s training data is non-English, including underrepresented languages like Swiss German and Romansh. Unlike commercial models, Apertus’s (from the Latin for “open”) architecture, model weights, training data, and development recipes are all openly accessible and fully documented, ensuring compliance with Swiss data protection laws and EU AI Act transparency requirements. The models are freely available under a permissive open source license for educational, research, and commercial applications." https://lnkd.in/e74P8pvZ
15
2 Comments -
Furu Wei
Microsoft Research Asia • 13K followers
Introducing Generative Adversarial Distillation (GAD): a novel GAN-style formulation and framework that facilitates both on-policy and black-box distillation of large language models (LLMs). GAD is the first technique to enable block-box on-policy distillation from proprietary teachers where internal logits or parameters are inaccessible, or distillation between teacher and student LLMs with incompatible vocabularies. GAD expands our prior work on white-box on-policy distillation (i.e., MiniLLM), pioneering block-box on-policy distillation for LLM training. Specifically, GAD frames the student LLM as a generator and trains a discriminator to distinguish its responses from the teacher LLM’s, creating a minimax game. The discriminator acts as an on-policy reward model that co-evolves with the student, providing stable, adaptive feedback. Experimental results show that GAD consistently surpasses the commonly used sequence-level knowledge distillation. In particular, Qwen2.5-14B-Instruct (student) trained with GAD becomes comparable to its teacher, GPT-5-Chat, on the LMSYS-Chat automatic evaluation. The results establish GAD as a promising and effective paradigm for black-box LLM distillation. Our team has been conducting fundamental research in knowledge distillation with wide adoptions across the industry. - MiniLM: We introduced multi-head attention distillation, establishing the most effective distillation method for BERT-style models. The open-source MiniLM models (e.g., 6x384) have become the most widely utilized small encoder models on the Hugging Face. - MiniLLM: Our proposed Reverse KLD is recognized as one of the most effective, de facto on-policy distillation approaches for modern LLM training, which has been widely used by Thinking Machines, Gemma, and many other teams and models. - BitDistill: We proposed BitNet Distillation to finetune off-the-shelf full-precision LLMs (e.g., Qwen) into 1.58-bit precision (ternary weights {-1, 0, 1}), achieving performance parity with the full-precision counterparts on specific downstream tasks. - GAD: The development of Generative Adversarial Distillation (GAD) now allows for black-box on-policy distillation, overcoming two major prior limitations: (1) Distillation from proprietary teachers where internal logits or parameters are inaccessible; (2) Distillation between teacher and student LLMs with incompatible vocabularies. https://lnkd.in/gMaP2c7w
125
2 Comments -
William Lindskog-Munzing
Flower Labs • 3K followers
What if training powerful AI models didn’t require centralizing data, compute, or control? 👇 That’s the core question behind the SPRIND - Bundesagentur für Sprunginnovationen Composite Learning Challenge, and we’re excited to share that CambridgeFlower (team University of Cambridge & Flower Labs) has been selected to move into Stage 2! Today’s AI training pipelines assume, centralized data, homogeneous hardware, and always-on connectivity. Composite Learning challenges this by enabling decentralized, federated training across heterogeneous and real-world systems; from edge devices to data centers, across organizations. Over the next 12 months, with SPRIND - Bundesagentur für Sprunginnovationen’s support, we’ll be focusing on Federated & decentralized learning at scale, and robust training under heterogeneity and intermittency. 📣 Huge congratulations to the other Stage 2 teams exalsius (Alexander Acker, Soeren Becker), SymphonyLearn (PanocularAI, Arya Mazaheri), AETHER (Tilmann Bartsch, SEMRON), and Planetary Compute Platform (Phil Rohr, DeltaWave). And a big thank you to SPRIND - Bundesagentur für Sprunginnovationen and the jury -- Johannes Otterbach, Katharine Jarmul, Wolfgang Stille, Hans Ramsl, Gerhard Wunder, Andreas Unseld and Tian Li -- for supporting ambitious, systems-level AI research. 👉 If you’re building or thinking about federated, decentralized, or edge AI systems; follow along or reach out.
45
-
Bo Wang
Xaira Therapeutics • 21K followers
Great discussion in our lab today, thanks to Alhusain Abdalla , on Recursive Language Models (RLMs) and how they challenge our assumptions about long-context reasoning. A common belief in the LLM community is that long-horizon tasks are primarily a context window problem: if models could simply “see more,” they would reason better. In practice, we increasingly observe the opposite—context rot—where performance degrades as information density grows, even in frontier models. 🧠 RLMs propose a different framing Rather than forcing models to ingest ever-larger contexts, RLMs treat context as something to interact with programmatically. Conceptually, this represents an evolution: • RAG is efficient but inherently lossy—if retrieval misses the right evidence, the model fails. • Agentic loops improve flexibility, but remain linear and constrained by short-term memory. • Recursive Language Models take a more radical step: the model writes code, recursively decomposes the data, spawns sub-calls, and aggregates results—without ever needing to load the full corpus into its own context window. 🔑 The key insight is that RLMs decouple reasoning capability from context length. The root model never sees the entire dataset—only structured outputs from recursive sub-tasks—enabling scalable reasoning over effectively unbounded data while keeping inference focused and efficient. We’re currently experimenting with their code on several clinical trial–related use cases, and we’re excited to share more results soon 🔥 This feels like an important step toward reasoning systems that scale not by reading more, but by thinking more structurally.
101
4 Comments -
Vinod Suresh Ph.D
CitiusTech • 2K followers
Beyond Transformers: What’s Next for LLMs in the Deep Learning Era? In just a few years, large language models (LLMs) have gone from research curiosities to powering everyday tools — from coding assistants and marketing copywriters to legal advisors and medical copilots. Under the hood, deep learning has driven this revolution — particularly the Transformer architecture. But what if we’ve only scratched the surface? The Transformer Era: Brilliant but Limited? Transformers brought us the superpowers of: Parallel training over large datasets, Attention mechanisms that let models weigh context efficiently, And scaling laws showing that “bigger is better” — at least until recently. Yet, we’re also seeing diminishing returns: GPT-4’s leap over GPT-3 is real, but not revolutionary. Training and inference costs are skyrocketing. And many LLMs still hallucinate, lack reasoning, and struggle with real-time constraints. It’s time to ask: What comes after Transformers? Neurosymbolic LLMs: Logic Meets Language A new wave is brewing — hybrid architectures that blend deep learning with symbolic reasoning. Imagine LLMs that: Not only generate fluent text, but also verify facts logically, Can explain their decisions, not just output tokens And act like reasoning agents rather than black boxes. Think OpenCog, DeepMind’s Gato, or Meta’s Toolformer as early signals. Memory-Augmented LLMs: From Static to Stateful The next generation of LLMs may be stateful and evolving, not just stateless text generators. New research shows promise in: External memory modules (like RETRO, RAG, or MemGPT), Long-context reasoning across multiple sessions, And lifelong learning — updating models without retraining from scratch. This makes LLMs more like digital companions than just chatbots. LLMs as Cognitive Engines, Not Just Models Here’s a new perspective: LLMs are not just NLP models anymore. They’re cognitive frameworks — bridging perception, memory, reasoning, and action. What if we start designing LLMs as: Multimodal agents (text, image, speech, video), World simulators (for gaming, robotics, strategy), Or even scientific assistants that explore hypotheses and automate literature review? Call to Builders: Don’t Just Use LLMs — Reinvent Them As engineers, researchers, and entrepreneurs, we’re entering a rare moment: A new “post-transformer” paradigm is forming. Ask yourself: What if we mix symbolic AI with LLMs? What if models verify instead of just generate? What if context is not just tokens, but structured memories? Final Thought LLMs have brought us far, but the next leap won’t come from just adding more parameters. It’ll come from rethinking the architecture, memory, reasoning, and embodiment. Let’s not wait for someone else to define what’s next. Let’s build it. #DeepLearning #LLMs #AIResearch #SymbolicAl #NeuroSymbolic #CognitiveAl #MLEngineering #FutureOfAl #LinkedInAl
6
-
Marcel Salathé
École polytechnique fédérale… • 37K followers
After years of building and refining #MyFoodRepo 📱 for our own nutritional studies at EPFL's Digital Epidemiology Lab, we're making it available to researchers everywhere. If you've ever run a study with dietary data, you know the pain: food frequency questionnaires that rely on faulty memory, manual food diaries that cause dropout, and research-grade apps that work in a pilot but frustrate participants at scale. Commercial apps might seem tempting, but they're often privacy-invasive, and you never really know if you'll get the data you need, or under what conditions. We built something different. Participants snap a photo, scan a barcode, or type what they ate in plain language. Cutting-edge AI, i.e. the same vision and language models powering today's leading AI applications, identifies foods, estimates portions, and maps to nutritional databases in real time. No endless dropdowns. No clunky search interfaces. And you get full access to all the raw, timestamped data. Privacy is built into the architecture, not bolted on as an afterthought. We collect zero personal information from participants, no names, no emails, no phone numbers, no location data. We don't know who your participants are, because we neither want to nor need to. Over a dozen research groups across multiple countries already use MyFoodRepo, from clinical trials at university hospitals to large-scale digital cohorts with over 1,000 participants. The platform has generated hundreds of thousands of annotated meal entries and contributed to peer-reviewed publications. It works because we made it easy enough that people actually stick with it. Our own studies depend on it. Now you can use it too. You get your own isolated cohort, real-time data monitoring, automatic nutritional mapping to official databases or your own custom food lists, and full data export via API or CSV. For academic research, it's typically free through our collaboration model. If you're planning a clinical trial, digital cohort, or any study involving dietary data, this might be what you've been looking for. 👉 https://lnkd.in/eraiPNf3
136
9 Comments -
James Lindsay
Dana-Farber Cancer Institute • 745 followers
New paper from Yann LeCun. It's a great read as always and it comes out at the right time when I think thousands [of software engineers] are really getting into their coding agents and discovering glimpses of their static nature (forget to save a session?). I think part of the art of writing a high-impact paper is generating a reaction. I can't help but feel like the threat of a biologically inspired engineered optimization algorithm is more for the dramatic story telling. https://lnkd.in/eZAuqWkN
5
-
Jonas Wengel
Jonas Wengel - Consulting • 2K followers
AI built for society, not for profit. 💡 ETH Zurich and EPFL are about to make a bold move. They are launching a language model without chasing profit. Built for the public, researchers, and education. It’s designed with transparency, data privacy, and long-term public interest in mind. We’re used to seeing Big Tech race ahead with closed systems. Now we have something in Europe which is more open. Some interesting facts: It’s hosted on Swiss infrastructure for data sovereignty. 🇨🇭 It’s trained on publicly available, licensed data, not user content. ETH wants to invite universities, authorities, and civil society to co-develop and explore it together. They want to understand these models instead of just using them. I love that. More of this mindset, please. What do you think? Do we have a chance to lead in ethical, public-focused AI? -J Content: Jonas and AI
5
-
Lingkai Kong
Harvard University • 2K followers
You may already be amazed by diffusion models for generating images and videos—but did you know they can also help wildlife conservation🐘? We propose DiffOracle, combining diffusion models & game theory to stop poaching! Arxiv: https://lnkd.in/ehkwF6YJ We use a conditional diffusion model to forecast the number of snares in each grid cell of the national park, capturing complex high-dimensional distributions. On poaching data from Murchison Falls National Park (Uganda), our model outperforms baselines in forecasting accuracy. To account for imperfections in the learned diffusion model, we formulate a robust patrol optimization problem as a two-player zero-sum game between a ranger defender and a nature adversary that selects the worst-case poaching activity within a constrained space. Empirically, we find that our game-theoretic approach yields patrol strategies with significantly lower regret than baseline methods, resulting in more effective wildlife protection in national parks. Thanks for all my wonderful collaborators including Haichuan Wang, Yuqi Pan, (Adam) Cheol Woo Kim, Mingxiao Song, Alayna Nguyen, Tonghan Wang, Haifeng Xu and Milind Tambe #DiffusionModel #GenerativeAI #Conservation #Sustainability #ClimateChange
52
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content