It is one reason why I think the push for smaller, local models is more complicated than people think. If you want good answers, especially good answers to unexpected problems, frontier models will generally outperform smaller models on a wide range of tasks.
RESEARCH
-
RDUs for Continuous Inference and Large-Scale AI Workloads
By
–
Built differently for inference 🦾
— SambaNova (@SambaNovaAI) 11 mai 2026
RDUs are designed to keep execution streaming continuously across the system, from memory to compute to parallel execution at scale.
256 RDUs working together can generate thousands of tokens in parallel for modern AI workloads.
Learn more:… pic.twitter.com/2Rf4GTon3cBuilt differently for inference RDUs are designed to keep execution streaming continuously across the system, from memory to compute to parallel execution at scale. 256 RDUs working together can generate thousands of tokens in parallel for modern AI workloads. Learn more:
-
Anthropic cautious vs OpenAI open on cyber-capable AI releases
By
–
Anthropic: “Claude Mythos is too cyber-capable to release broadly. We need tight controls. 😳”
— Chubby♨️ (@kimmonismus) 11 mai 2026
OpenAI: “Here’s GPT-5.5-Cyber, Codex Security, Trusted Access tiers, repo scanning, patch generation, and red-team workflows. Please be verified first, but yes, go find the bugs. 😎” https://t.co/8vZwl8tVwfAnthropic: “Claude Mythos is too cyber-capable to release broadly. We need tight controls. ” OpenAI: “Here’s GPT-5.5-Cyber, Codex Security, Trusted Access tiers, repo scanning, patch generation, and red-team workflows. Please be verified first, but yes, go find the bugs. ”
-
Sharpness-Aware Pretraining for Mitigating Catastrophic Forgetting in AI Models
By
–
“Sharpness-Aware Pretraining Mitigates Catastrophic Forgetting” Better base models don’t always become better fine-tuned models, because post-training can overwrite pretrained capabilities. So this paper pretrain models into flatter loss regions, so later updates from
-
Manifold Steering for Neural Network Representation and Behavior
By
–
"Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior" Most activation steering treats concepts like directions, but model representations are often curved, so straight line steering can cut through unnatural internal states. The key idea
-
The Evolution and Challenges of AI Coding Agents
By
–
The question has moved past whether coding agents can produce code that works. The harder question is whether they can complete real software work safely, measurably, and repeatedly — and how we supervise it. Great piece by @realjustinbauer on the evolution of coding agents and
-
Critical reason to open up about AI use in academia
By
–
This seems like a critical reason to open up about AI use in academia. Scholars are using old AI models, badly, and not talking about it. New models hallucinate very few citations, and good agentic harnesses drop that further. Being open about use would help us make new norms.
-
AI Consciousness: Expert Views on LLMs and Biological Replication
By
–
I wrote about the idea of AI consciousness for @WIRED a while ago. Most experts think LLMs are to replicate biological consciousness (at least, according to our best understanding of it). But some eminent scientists are trying to build new kinds of AI systems that could, perhaps,