China just isn't that AGI-pilled, no matter how much we want (or need) them to be to justify what we want (or need) to build for ourselves
(wrote this more than two years ago)
https://t.co/c2ECW7PnLx
My full talk on the future of AI & media is up!
I used @alexolegimas's prompt of "What will be scarce?" to propose 4 ways that media is changing, and how writers can still win in the AI age:
1) Secrets > summaries
Reporting is the act of taking private knowledge and making it public: when you get a source to tell you about corporate malfeasance, or venture to a remote town that few people have been to, or sneak your way into an underground party, you are working in a space where there is no training data.
2) Live interaction > static content
We’re not far from a world where AI can replicate any prose style. But readers want to know there's a real person generating the text—not just the final presentation, but the proof of work behind it. For creators, doing live events, podcasts, and meetups reveal the life behind the voice. And if I care about my ideas, I want people to know about them, no matter the format.
3) Founders > bureaucracies
AI is already allowing startups to run leaner by helping founders act as their own marketer, data scientist, engineer, etc. It's the same in media — AI is a boon to jacks-of-all-trades. There’s a lot of stuff AI does that I don't want to: verifying cites, reading contracts, negotiating speaking fees. It's an amazing time for independent creatives who want to direct their own vision.
4) Personal style > polish
The house style in most newsrooms is extremely LLMable. What stands out (besides reporting) is a distinct and authentic first-person voice, even if that means the occasional typo / provocation / admitting "I'm not really sure." After all, trust isn't about the perfect sentence: it’s about the track record of who says it. And the stronger your brand, the more trusted you’ll be.
I spend a lot of time covering the real disruptions AI brings. But I also believe, for those with the gumption to seize the opportunity, there's never been a better time to be a writer 🧡
AI factories are on track to break the global power grid. ⚡️
In this clip from GTC, @vsiv explains how @EmeraldAi_ is solving the energy crisis by turning data centers into "flexible power users."
By dynamically scaling power consumption on demand, they are pulling off a massive triple-win:
- Allowing AI factories to scale faster
- Keeping local community energy bills low
- Keeping the electric grid stable
The future of AI isn't just about faster chips—it's about smarter energy.👇 Find the full conversation in the replies!
My time at Ai2 / @allen_ai has come to an end.
Ai2 is a wonderful place. The last 2.5+ years building Olmo, Tulu, and other projects will be one of the peaks of my entire career. I'm extremely thankful for my teammates and the open community who made this work possible.
For me, it's time to try something different. I will still be working in the open model & open science spaces (more news on that soon). In the meantime I'll be spending a few months learning, chatting with a broader network, getting married (!!) and most importantly recharging from pouring my soul into this place.
I've attached the note I shared with the team and some fun photos from our time together. I'll keep cheering for Ai2 and am excited to see what you build next.
every evals/analytics startup is going through a onetime generational upgrade into a continual learning platform in 2026
many will fail but as always the tasteful ones win
🆕Grok Imagine’s Video Agent Moment: Cosmos, xAI, World Models, Generative UI, & the Codex Phase for Video!
https://t.co/vYADYniTj2
@EthanHe_42, former @xai world model lead and @nvidia Cosmos researcher, explains why AI video may follow the same path as coding agents, how Grok Imagine went from zero to one, why text-to-video is only the autocomplete phase, how world models become real-time and interactive, why language models may become the control layer for video, and why the future of AI video may look less like a prompt box and more like an agent with a camera, editor, timeline, and tool belt.
AI training was just the warm-up act. The real infrastructure war is being fought in inference.
According to Val Bercovici (@weka) inference is rapidly fragmenting into 3 distinct compute tiers:
1️⃣ Voice Agents (Ultra-low, fixed latency to avoid awkward gaps)
2️⃣ Chat/Research Sessions (Standard mid-tier flows) 3️⃣ Agent Swarms (Multi-day parallel batch sessions)
If you can cut the latency on a multi-turn agent swarm, you eliminate massive compounding delays. This is completely rewriting the rules of AI infrastructure.
Val broke down the future of compute at GTC earlier this year.👇
AI agents are powerful, but blindly replacing humans is a recipe for disaster.
This isn’t about full autonomy—it’s about out how to let agents "run free" without breaking things.
Travis Perkins from @Tailscale drops some wisdom on balancing speed, guardrails, and human-in-the-loop innovation at #NVIDIAGTC👇
The cloud-hosted AI hype is missing the bigger picture.
The real paradigm shift? The inevitable move to on-prem model hosting, paired with economically viable labor robotics.
When robots start taking over the manual, 4 AM shifts humans don't want to do, the entire cost structure of our economy gets rewritten.
@wschenk breaks down the future of AI and robotics - full interview on YT.
I was in China a couple of weeks ago with @natolambert , @xeophon, the @readsail team including @caithrin, @azeem, @jasminewsun, @kevinsxu, and other wonderful humans.
Over dinner in Shanghai, a friend asked:
“Are these robot dogs and humanoids actually running large language models? How does a language model… move a leg?”
That question became the starting point for my new piece as I processed how Chinese and US robotics industries are progressing and where they are excelling.
🆕Daytona’s Agent-Native Compute: 60ms sandboxes, 50K startups in 75 sec, 850K daily runs, RL/evals, CLI > MCP, & the end of localhost https://t.co/ceZgmkIFqe
@daytonaio CEO @ivanburazin explains why AI agents need composable computers, how Daytona pivoted from human dev environments to agent sandboxes, why bare metal and stateful snapshots matter, how RL workloads went from 0% to ~50% of usage, why Kubernetes breaks down at agent scale, and why the AI cloud may look more like Stripe than AWS.
We just wrapped what began as an 8-hour challenge - and it ran for 200 hours without a failure
Shoutout to the team for the hardcore engineering behind F.03 and the robust Helix models powering it