World Wire

Evaluating AI models trained on brain signals has long been a messy, inconsistent topic. Different research groups use different preprocessing pipelines, train models on different datasets, and report results on a narrow set of tasks — making it nearly impossible to know which model actually works best, or for what. A new framework from Meta AI team is designed to fix that. Meta Researchers have released NeuralBench , a unified, open-source framework for benchmarking AI models of brain activity. Its first release, NeuralBench-EEG v1.0 , is the largest open benchmark of its kind: 36 downstream tasks, 94 datasets, 9,478 subjects, 13,603 hours of electroencephalography (EEG) data, and 14 deep learning architectures evaluated under a single standardized interface. https://ift.tt/yCqbYBP The Problem NeuralBench Solves The broader field of NeuroAI where deep learning meets neuroscience has exploded in recent years. Self-supervised learning techniques originally developed for...

Training frontier AI models is not just a compute problem — it is increasingly a networking problem. And OpenAI just introduced its solution. OpenAI announced the release of MRC (Multipath Reliable Connection) , a novel networking protocol developed over the past two years in partnership with AMD, Broadcom, Intel, Microsoft, and NVIDIA. The specification was published through the Open Compute Project (OCP), enabling the broader industry to use and build on it. Why Networking is the Hidden Bottleneck in AI Training To understand why MRC matters, you need to understand what happens inside a supercomputer during model training. When training large AI models, a single step can involve many millions of data transfers. One transfer arriving late can ripple through the entire job, potentially causing GPUs to sit idle. Network congestion, link, and device failures are the most common sources of delay and jitter in transfers — and these problems get more frequent, and harder to...

Zyphra AI has released ZAYA1-8B, a small Mixture of Experts (MoE) language model with 760 million active parameters and 8.4 billion total parameters. Trained end-to-end on AMD hardware, the model outperforms open-weight models many times its size on math and coding benchmarks, and is now available under an Apache 2.0 license on Hugging Face and as a serverless endpoint on Zyphra Cloud. With under 1 billion active parameters, ZAYA1-8B achieves scores competitive with first-generation frontier reasoning models like DeepSeek-R1-0528, Gemini-2.5-Pro, and Claude 4.5 Sonnet on challenging mathematical reasoning tasks. With its novel test-time compute methodology called Markovian RSA, it surpasses Claude 4.5 Sonnet and GPT-5-High on HMMT’25 (89.6 vs 88.3) and closes in on frontier open-weight models like DeepSeek-V3.2 on mathematics benchmarks. What is a Mixture of Experts Model and Why Does Active Parameter Count Matter? The distinction between ‘active’ and ...

In this tutorial, we build a Groq -powered agentic research workflow that runs directly using Groq’s free OpenAI-compatible inference endpoint. We configure LangChain’s ChatOpenAI interface to work with Groq by setting the Groq API key and base URL, allowing us to use fast hosted models such as llama-3.3-70b-versatile for tool-based reasoning. We then connect the model with practical tools for web search, webpage fetching, file handling, Python execution, skill loading, sub-agent delegation, and long-term memory. By the end of the tutorial, we have a working Groq-based multi-step agent that can research a topic, delegate focused subtasks, generate structured outputs, and save useful information for later runs. Copy Code Copied Use a different Browser import subprocess, sys def _pip(*a): subprocess.check_call([sys.executable,"-m","pip","install","-q",*a]) _pip("langgraph>=0.2.50", "langchain>=0.3.0", "langchain-...

Most agentic applications today have a memory problem. Every time a user opens a new session, the agent starts from zero. There is no recollection of what was discussed, what workflows were in progress, or what decisions were already made. The session ends, and everything disappears. For dev teams shipping production agentic applications, the only way around this has been to hand-roll a storage layer from scratch, picking a database, serializing state, managing session IDs, and connecting it back into the agent runtime before writing a single line of actual product logic. The Enterprise Intelligence Platform by CopilotKit solves this by providing a managed infrastructure layer that handles state and memory automatically.  It works independently of the agent framework – any agent can have memory. Give CopilotKit a on GitHub What is CopilotKit Intelligence? CopilotKit is the frontend stack for AI agents — production infrastructure for building Generative UI ...

There is a particular kind of irony that the legal profession rarely gets to witness in such pristine form. In May 2025, Latham & Watkins a firm that routinely bills over $2,000 an hour for its partners and counts Anthropic among its clients filed a court declaration in Concord Music Group v. Anthropic that contained fabricated citation details. The citations weren’t invented by a sleep-deprived associate pulling an all-nighter. They were generated by Claude, the very AI model that Latham & Watkins was in court defending. Sit with that for a moment. The lawyer arguing that Claude is not a copyright infringement machine used Claude to format a legal citation in an active case and Claude got the authors wrong, the title wrong, and nobody caught it until opposing counsel started digging. The irony isn’t just delicious. It’s instructive. Because what happened inside that filing is a near-perfect X-ray of the structural problem that AI poses for legal prac...