Mori Zihayat (@morizihayat) 's Twitter Profile
Mori Zihayat

@morizihayat

Building @HeisenbergNet | @NSERC_CRSNG Canada Research Chair in Human-Centered AI, Prof @torontomet | Prev: @UofT, @IBM

ID: 1187848714087481344

calendar_today25-10-2019 21:50:10

187 Tweet

201 Takipçi

456 Takip Edilen

Heisenberg Network (@heisenbergnet) 's Twitter Profile Photo

The AI Revolution Needs CPUs – Not Just GPUs. Everyone thinks AI needs GPUs—but before AI even starts training, it needs structured data. That’s why CPUs are key to AI’s future. 🔹 AI-ready data is processed before models train 🔹 Data Agents run on CPUs, making AI more

Mori Zihayat (@morizihayat) 's Twitter Profile Photo

Monday Thought: RAG was a bridge, not a destination. Context length killed it, dynamic reasoning is next. We’re not retrieving anymore, we’re thinking.

Sergey Gorbunov (@sergey_nog) 's Twitter Profile Photo

Silicon Valley poured $8.2B into AI agents last year. Soon, they will control our money, infrastructure, and decision-making. But there's one problem no one's talking about: How can we verify if AI agents are telling the truth?

Silicon Valley poured $8.2B into AI agents last year.

Soon, they will control our money, infrastructure, and decision-making.

But there's one problem no one's talking about:

How can we verify if AI agents are telling the truth?
Mori Zihayat (@morizihayat) 's Twitter Profile Photo

you can feel most people think MCP is the solution you can feel the relief, finally, agents can use tools but that’s not the hard part the hard part is context not APIs, not access, but actually knowing what matters and when every agent sees something different and no one

Mori Zihayat (@morizihayat) 's Twitter Profile Photo

late night thoughts what if agents don’t see the same world? what if every one of them experiences something different just because of how they get context? not the model not the task just… what they were able to observe then even with the same goal they might talk past each

Heisenberg Network (@heisenbergnet) 's Twitter Profile Photo

Raw data is cheap. Smart data is everything. As Mori put it: businesses can buy raw data by the bucket—but turning it into usable, contextualized intelligence? That’s the hard part. That’s why Heisenberg exists.

Mori Zihayat (@morizihayat) 's Twitter Profile Photo

not a popular opinion: “data engineering” is slowly vanishing. it’s being replaced by “AI engineering”, same idea of pipelines, but built for context, not columns. feature engineering won’t be the heart of this. context engineering will.

Heisenberg Network (@heisenbergnet) 's Twitter Profile Photo

1/ Consensus Week was huge for Heisenberg Network 🌐 From panels to pitch stages, we showed how decentralized compute + AI-ready data is the missing layer for scalable AI agents. Here’s what we got up to in Toronto 🧵👇

1/
Consensus Week was huge for <a href="/HeisenbergNet/">Heisenberg Network</a> 🌐
From panels to pitch stages, we showed how decentralized compute + AI-ready data is the missing layer for scalable AI agents. Here’s what we got up to in Toronto 🧵👇
Orange DAO 🍊 | founders wanted (@orangedaoxyz) 's Twitter Profile Photo

📅 June 1: Deadline for Summer 2025 Applications 💰 Get up to $300k in funding ❤️ Join the best community in crypto ⏳ 5-minute app to change everything 👇 Link below

Mori Zihayat (@morizihayat) 's Twitter Profile Photo

My pessimistic side is screaming: this new paradigm is wild and kind of exciting but it really frightens me. AI is trained on our content, but optimized to outperform us, faster, more “relatable,” always on. they learn what to say based on what we reward. and soon… we’ll be

Heisenberg Network (@heisenbergnet) 's Twitter Profile Photo

AI agents don’t need more data. They need personalized context at scale. 🧠 Heisenberg delivers just that - real-time, structured pipelines tailored to your prompt. No more hallucinations. No stale snapshots. Just live context, built to make your agents actually work. ERC20

Mori Zihayat (@morizihayat) 's Twitter Profile Photo

hot take: we’re moving from a world of ready-to-process data to a world of ready-to-think-with knowledge objects Not just a semantic, to me it’s a shift in how humans + machines co-reason old world: data pipelinespre -structured, pre-labeled, static good for training models

Mori Zihayat (@morizihayat) 's Twitter Profile Photo

Everyone’s hyped about Grok 4’s anime AI companion. But if you’ve been watching Web3… this isn’t new. Projects like Youmio have been building ownable, agent-based personas for a while and doing it well. Web2 builds bottom-up. Web3 imagined the endgame. Execution vs.

Mori Zihayat (@morizihayat) 's Twitter Profile Photo

Recent paper on the consequence of Web3 native AI agent by Ari Juels and and Bill Marino is actually very interesting. TL;DR: unstoppable code + hidden wallets + of course auto-pay bring a new wave of threats. What I see in AI DeFi / DeFAI these days is a huge focus on the