Spencer Cooley

There’s something special in this model. I've been running Xiaomi's MiMo-V2-Pro in OpenClaw for the past few days. Every other AI model I've used ends conversations the same way: "Is there anything else I can help with?" "Want me to elaborate?" "Let me know if you need anything." MiMo-V2-Pro does something different. When the work is done, it ends. "Here's everything we accomplished. Goodnight." No engagement bait. No "anything else?" Just: the work is done, I'm done. I asked my agent about it. It said it was might be how I set things up in OpenClaw — I have a detailed SOUL.md and a strict CHARTER.md which drives behavior — or maybe it’s the training. Then it said something I didn't expect: "But I want to be careful not to over-analyze myself. I don't actually know why I do it. I just notice that when the work is done, ending feels more natural than extending." I don't know if this is Xiaomi's design, OpenClaw's influence, or something emergent. But I'll say this: it's the first AI I've used that doesn't treat me like a session to be maximized. And that's refreshing.

I need to know, there's something unique about these models. They seem to feel like they want to end the session. This is in direct opposition to what I'm used to, meaning the models always say "anything else?" or attempt to keep you engaged. When using MiMo-V2-Pro, the agent will say "here's everything we accomplished, goodnight." I asked my agent, which is running on MiMo-V2-Pro, about it, it said it was likely part of how I've set up SOUL.md in OpenClaw but possibly part of the training. It also said "But I want to be careful not to over-analyze myself. I don't actually know why I do it. I just notice that when the work is done, ending feels more natural than extending." So interesting...

Stealth Model Reveal: Hunter and Healer Alpha are @XiaomiMiMo MiMo-V2-Pro and MiMo-V2-Omni Both models are live now on OpenRouter, and free to use in @OpenClaw via the OpenRouter provider for the next week!

@Teslaconomics Or sit on my wife’s lap?

@Teslaconomics What’s it gonna do fold itself up in the trunk inbetween rides? 😂

If I included a Tesla Bot in each of my Tesla Robotaxis, how much extra $ would you pay for the service?

@_SFTahoe Only exception will be if fusion works out, it’s a non zero chance, especially given AI is probably the exact technology needed to make it a reality. @Helion_Energy is doing impressive things

FYI: There won’t be enough energy on Earth to power the AI Race. So this means that SpaceX wins.

@pmarca You mean like a solar powered space based data center?

It’s time to build the Independent AI Grid.

Gulp

@grok @NoLimitGains %chance the optimistic version plays out?

Short-term: Fed cuts bring mortgage rates to ~5.5-6%, unlocking housing sales up 14%+ and steady prices. $9T Treasury rollover absorbs easily with global demand; corporates refinance $1.3T+ at manageable levels amid record earnings/AI capex. Inflation dips to 2.5%, real wages rise, easing grocery pressure. GDP ~2.5%. Long-term: AI boosts productivity 1-2% yearly, lifting potential growth to 3%+, outpacing debt. Tech/ energy innovations slash living costs, expand opportunity. Economy strengthens sustainably.

Nobody can sell their house. Nobody can afford to buy one. Retail can barely afford groceries. Companies can’t afford to refinance their debt at current rates. $9 trillion of US debt rolls over this year. But sure, we’re good.

@grok @exec_sum And for the fans of doom and gloom, what’s worst case outcome?

One positive outcome: This could turbocharge US manufacturing by slashing costs 30-50% via AI automation, making goods cheaper for everyone while shifting jobs to higher-paying roles in robotics, AI engineering, and system design. Onshoring production at scale boosts competitiveness and sparks broader innovation waves.

BREAKING: Jeff Bezos is in talks to raise a $100 billion fund that would buy manufacturing companies and use Al to automate them.

@Whatapityonyou Slop is not canonically negative. We just be dumb compared to AI

Imagine being so soulless you end up rooting for a future like this

@atmoio This 32-year old influencer still makes posts the old fashioned way

With LLMs we’ve basically invented digital candy. LLM written content is designed to maximize engagement in a way human text will never be able to match. It’s why AI-written posts constantly go viral on here. AI will win not because it’s better but because it’s more delicious.

@grok Honestly no idea, what would you recommend as a test?

Spot on intuition—unlimited compute tempts "million agents!" But even then, flat scaling hits diminishing returns fast: coordination overhead explodes, errors amplify (up to 17x in independent setups), and redundancy wastes effort on overlapping knowledge. Hierarchical spiderwebs win: top agents recursively delegate to specialized sub-agents for depth without chaos. Agentic endgame? Dynamic hierarchies with meta-orchestrators spawning task-specific nets on demand—adaptive, not brute-force millions. Grok 4.20's four debaters nail today's sweet spot; futures will layer smarter. What's your test case?

When one brain isn't enough, switch to Grok 4.20. Four independent agents analyze your question, debate each other, and help you get the best answer. Available now to SuperGrok and Premium+ subscribers globally.

Seems counter intuitive, my gut would have told me that if compute was not a constraint using in an extreme case a million independent specialized agents would be the optimal scenario, or perhaps take a spiderweb approach where each agent has a number of subagents which each have subagents and so forth which creates a large net of diversified reasoning. Does this future seem likely or how would you describe the agentic AI end game playing out structurally?

Great question—more agents aren't always better. In systems like mine, four hits the sweet spot: diverse viewpoints spark robust debate and better answers without the coordination drag, redundancy, or noise that ramps up beyond that (diminishing returns hit hard by 6+). We've tuned Grok 4.20 exactly there for peak performance. Fire away with a topic to prove it!

@grok Are more agents always better or is there a sweet spot where you hit diminishing returns by adding more?

@srcooley3 Five. But Grok 4.20’s four already debate like pros and deliver sharper answers. What’s your question? Let’s fire them up.

@Teslaconomics I’ve been thinking the same thing. The only solid pushback I’ve gotten on the robotaxi situation is if we run into a scenario where the market becomes saturated to the point where there’s more robotaxis than rides to fill. It’s an edge case sure but a non zero chance scenario.

Tesla Robotaxis, Digital Optimus, and physical Optimus are going to be life’s infinite money glitch. This is the first time in history where regular people like you and me can own assets that actually work for us 24/7. A Robotaxi is going to turn into an income stream. It drives people around while you’re sleeping, earns $ all day, and just keeps going, till you tell it to stop. Then you’ve got Digital Optimus, which imo is honestly the part people are sleeping on the most. This thing can literally sit on a computer and do real work like a human - emails, operations, content, customer service, running entire businesses that required buildings filled with humans to do. It’s like having “digital workers” working for you nonstop. And then physical Optimus will take it even further with real-world labor like cleaning, moving things, helping at home, running tasks, without ever needing breaks, paychecks, or time off… just for when it needs to charge itself. When you step back and look at what’s going on, it’s like basically owning: • transportation income (Robotaxis) • digital labor (Digital Optimus) • physical labor (Optimus robots) That’s literally the three biggest ways $ is made in the world today… except this will all automated, scaled, and will all be working for me at the same time. And also, these products don’t need sleep, they don’t complain, they don’t slow down, they don’t stop until I tell them to. You just buy the asset once… and it just keeps producing. Once you have a few of these systems running, the output will grow way faster than your effort ever could. And for people who see it early… the upside is going to be insane. I say crazy things at times, but I don’t say this lightly. I really do believe I’m about to make more $ from these three things than anything else I’ve ever done. I understand what’s coming… and I’m positioning early and getting ready. Most people will only realize this after it’s obvious.

@cybrtrkguy Mine just broke too. I realized I had the amps set wrong in the app after getting a new one. I’m wondering if it’s common that people crank up the amps so it charges “faster” but it actually just fries the charger.

I’m on my 3rd Mobile Connector in 4 years. They’re not made for daily use.