From under 28% to above 50% in eight months on Terminal-Bench is a pace that's easy to miss when everyone's focused on frontier model releases. Open models closing the gap this fast changes the economics for a lot of production use cases.
RESEARCH
-
AI Resolves 54-Year-Old Arithmetic Geometry Conjecture
By
–
Resolving a 54-year-old conjecture in arithmetic geometry is the kind of AI result that actually moves the needle on whether these systems can do real mathematical reasoning. Benchmarks tell you about pattern matching. Open problems tell you about understanding.
-
HF Spaces Feature Enables Public Model Demos With Protection
By
–
Smart feature. The "I want a public demo but don't want people cloning my model" problem has been a real friction point for researchers. Combined with custom domains this makes HF Spaces genuinely production-viable.
-
OpenSage: The First Self-Designing Agent Development Kit
By
–
1/ Introducing OpenSage: the first AI-centric Agent Development Kit (ADK). Today's agents are designed by humans — fixed topologies, handcrafted tools, rigid memory. It's time for agents to design themselves. [Translated from EN to English]
→ View original post on X — @berkeley_ai, 2026-03-23 15:27 UTC
-
Moving Toward Race-Neutral Clinical Algorithms in Healthcare
By
–
We have a new piece in Nature Health led by @dmshanmugam, @sidhikab1, and a wonderful team of coauthors on how to move towards a world in which race is not used in clinical algorithms! Divya Shanmugam (@dmshanmugam) New in Nature Health: how might we move towards a world in which race is not used in clinical algorithms? We need (1) careful comparison of race-aware and race-neutral algorithms and (2) systemic efforts to address underlying disparities. — https://nitter.net/dmshanmugam/status/2036097222404628670#m
→ View original post on X — @berkeley_ai, 2026-03-23 15:11 UTC
-
Physics-Informed Neural Operators: Learning General PDE Solvers
By
–
Physics-Informed Neural Operators: Learning The Solver, Not Just One Solution
— Mathelirium (@mathelirium) 23 mars 2026
Our PINN scene learned one solution field for one PDE setup. A Physics-Informed Neural Operator learns the map from input fields, like material coefficients or source terms, to the full solution across… pic.twitter.com/bAtfX2ashePhysics-Informed Neural Operators: Learning The Solver, Not Just One Solution Our PINN scene learned one solution field for one PDE setup. A Physics-Informed Neural Operator learns the map from input fields, like material coefficients or source terms, to the full solution across a whole family of PDE problems. So, the goal is no longer just one approximate answer, but a reusable solver-shaped object guided by the physics itself.
→ View original post on X — @animaanandkumar, 2026-03-23 14:45 UTC
-
MM-Zero: Building Multimodal AI Systems From Scratch
By
–
MM-Zero: Learning Multimodal Intelligence From Scratch
— Satya Mallick (@LearnOpenCV) 23 mars 2026
In this episode of Artificial Intelligence: Papers and Concepts, we explore MM-Zero, a new approach to building multimodal AI systems that learn from scratch without relying heavily on pretraining from separate models.… pic.twitter.com/VvyLNn86BeMM-Zero: Learning Multimodal Intelligence From Scratch In this episode of Artificial Intelligence: Papers and Concepts, we explore MM-Zero, a new approach to building multimodal AI systems that learn from scratch without relying heavily on pretraining from separate models.
-
The Evolution of AI: From Cloud to Embedded Physical Systems
By
–
Right now AI = centralized cloud. Text in, text out.
Massive models.
Huge data centers. But that’s the first chapter. We’re moving toward: → Multimodal AI
→ AI agents
→ Physical AI embedded in devices That changes everything. Because now AI depends on real-time data, not -
ICML 2026 removes reviews from LLM-using reviewers
By
–
As co-comms chair of ICML 2026 (w @kgorman), I'm super proud of how transparent we've been able to be on all of the (bold!) decisions made. Thanks to all the organizers (esp PC chairs) for being aligned on this. The community deserves to understand these important decisions ICML Conference (@icmlconf) To ensure compliance w peer-review policies, ICML has removed 795 reviews (1% of total) by reviewers who used LLMs when they explicitly agreed to not. Consequently, 497 papers (2% of all submissions) of these (reciprocal) reviewers have been desk rejected Details in blog post 👇 — https://nitter.net/icmlconf/status/2034279558598242523#m
→ View original post on X — @thegautamkamath, 2026-03-23 13:54 UTC
-
Atlas Robot Demonstrates Advanced Physical Capabilities and Skills
By
–
From Walking to Heavy Lifting: Atlas Shows Off Next-Level Skills
— Ronald van Loon (@Ronald_vanLoon) 23 mars 2026
by @lukas_m_ziegler
#Robotics #ArtificialIntelligence #Innovation #MI #ML #Tech pic.twitter.com/M6ofV5SRliFrom Walking to Heavy Lifting: Atlas Shows Off Next-Level Skills
by @lukas_m_ziegler #Robotics #ArtificialIntelligence #Innovation #MI #ML #Tech