Pathway

Pathway is a Bay Area based foundational AI research company. It is introducing the first post-transformer LLM that adapts in real time and reasons like humans. Alongside the model, Pathway has built the world's fastest data processing engine that supports various stages of an enterprise ready agentic pipeline. This year the Pathway High Prep problem raises the bar. It calls for mindful financial agents that solve specific hard problems and are ready for real world deployment. #Pathway #PostTransformer #AIInnovation #AgenticAI #EnterpriseAI #FutureOfIntelligence #interiittechmeet14 #interiit #techmeet

As Forbes’ Victor Dey writes, Pathway's BDH “may have sparked the beginning of a new era in AI — one where machines don’t just imitate the brain, but begin to think like it.” "The implications could be both technical and economic. Retraining large models costs companies billions each year in computing power and energy. A system that learns continuously could make AI development cheaper, faster, and more sustainable. Because the architecture keeps critical data close to its processing cores, it reduces latency and slashes compute costs." Our breakthrough post-Transformer architecture (#BDH) shows that AI can now self-evolve and reorganize itself, inspired by how our biological neurons learn and adapt. It’s a glimpse into what the next generation of intelligence should look like. We’re thrilled to see this conversation live — and even more excited about what comes next. Read the full piece below ⬇️ #AI

Curious about AI, neuroscience, or the future of ML architectures? Adrian Kosowski (CSO & co-founder of Pathway) is exploring with Jon Krohn Pathway’s new post-transformer architecture, #BDH that aims to bridge the gap between transformer models and how the brain actually works. In this SuperDataScience episode, learn about: 🧠 How we are rethinking attention in a way that’s more biologically plausible 🧠 Sparse activation (only ~5% of neurons “fire” at a time), which is closer to how the human brain operates, and much more efficient than dense attention in traditional transformers 🧠 How this architecture enables better reasoning over longer sequences, and more human-like generalization over time 🧠 And more benefits, including predictability, composability, interpretability and how some neurons can “specialize” for specific concepts,… Curious about AI, neuroscience, or the future of ML architectures? https://lnkd.in/emdNs3t7

🚨 OpenAI’s Big Claim: AI Isn’t Killing Coding Jobs, It’s Making Them Better In Q1 2025 alone, more than 76,000 jobs were lost to automation, and nearly 40% of employers expect cuts where AI can take over. With the rise of agentic AI, the fear is real: if coding assistants can debug and refactor code in seconds, what role is left for human developers? 💡 But OpenAI sees it differently. With its new GPT-5 Codex, the company argues coding isn’t dying, it’s evolving. Instead of replacing developers, Codex promises to accelerate creativity, reasoning, and problem-solving. 🧠 “I see new graduates on my team picking up programming at a speed I haven’t witnessed before,” Thibault Sottiaux, engineering lead for Codex at OpenAI, told me. “If you look at today’s AI tools, it’s clear they’re far from perfect. There’s still so much capacity in the world to absorb better, more powerful, and more delightful software.” 📊 Benchmarks back the claim. On SWE-bench Verified, GPT-5 Codex solved over half of code-fixing challenges, up from roughly one-third for its predecessor. 📌 OpenAI isn’t the only big player arguing that AI isn’t the IT job killer many feared. “Every minute, five to eight people sign up for a generative AI class on our platform,” Hugo Sarrazin, CEO of Udemy, told me. “Designing code requires critical thinking, which is fundamentally a human trait. Of course, AI will generate a lot more software, but you still need analysis, judgment, and testing. ⚔️ But skeptics like Jan Chorowski, CTO at Pathway (ex-Google Brain), remain unconvinced. “The key differentiation we have today, as humans, is the ability to come up with new ideas grounded in context,” he said. “AI lacks such contextualized judgment. Changing this is the challenge for the next decade.” 👉 Read my full analysis on Fast Company: https://lnkd.in/dddY77P5 #AI #OpenAI #Codex #FutureOfWork #AgenticAI #GPT5 #SoftwareEngineering #CodingJobs #Udemy

I am extremely proud to announce BDH - our post-transformer LLM architecture. It is a paradigm shift which opens the door to solving one of the last big challenges for AI - generalization over time. It offers strong theoretical foundations, interpretability, and performance which is competitive with Transformer. With this milestone, we have gained understanding of how intelligence emerges from neuron-to-neuron interactions, following the "equations of reasoning". BDH is the pivotal moment for AI that reasons and adapts like humans. Read the paper: https://lnkd.in/ewXatPq2 Congrats to the entire Pathway Team, and the co-authors: Adrian Kosowski, Jan Chorowski, Przemysław Uznański, Michał Bartoszkiewicz #AI #breakthrough

#1 on Hugging Face Trending Papers & #1 Paper of the Day. Meet Pathway's BDH — post Transformer architecture for LLMs that generalize, continually learn in real time, and are interpretable. #LiveAI #Pathway #BuildingLLMs

Today, Pathway is launching a new “post-transformer” architecture, Baby Dragon Hatchling (BDH), that paves the way for autonomous AI. Our research paper, The Missing Link Between the Transformer and Models of the Brain, tackles several of AI’s biggest challenges: generalization over time, learning in real time, and interpretability. Read on arXiv: https://lnkd.in/dNMSs46e

Why do some document AI projects for critical use cases deliver outcomes while others never evolve from the pilot phase? Join us for this live webinar by LandingAI and Pathway, as we unpack what actually matters. We'll also unveil releases that are reshaping how enterprises approach GenAI. Register below by marking "Attend" 👇

LLMs do not know what time it is. My opinion piece → https://lnkd.in/eBtD8Z_d Ask an ops agent: “How many shipments are delayed right now?” “Historically, the Frankfurt hub handles 12,000 to 15,000 packages during peak hours.” Every enterprise AI team has a version of this question. The one that breaks their RAG system. Not because it is complex. Because it contains the word “now”. Your RAG indexes documents. Great. But when did it last reindex them? Your LLM fetches the context via indexes. Perfect. But that context is stale. I keep coming back to this because it is the difference between AI demos and AI that actually works in production. The solution is not magical. It is a small, incremental improvement in the architecture that is easily replicable. MCP servers plus real-time pipelines. The full breakdown is in the Medium article. Your AI should know what changed five minutes ago. Otherwise, you are just running an improved search engine.