SmithsVision @Smiths_Vision

Humanoids excel in free space but struggle with real-world contact. Meet SceneBot 🤖 the first unified RL framework for ALL free-space locomotion, terrain traversal and object interaction! By conditioning on per-link contact labels, it masters complex, interaction-rich tasks like carrying a box upstairs. Code & data open-sourcing soon! 📦 🪜 Paper: arxiv.org/abs/2606.27581 Website: ericcsr.github.io/scenebot/

After years since launch and $95K in revenue FINALLY... added breadcrumbs to niche job board pages 😬 took me like 1 minute!!! why, we, founders sometimes are like this?

6-years later, nice to know something’s endure in this time of vast amounts of change…

Luke W. @Luke4OSS

6 years ago

This took patience

1 1 33 0 0

The business owners who struggle the most are the ones with companies that can only function when they’re in the room. That’s why we run on EOS at Headway, and the most valuable thing it gave us was the discipline to build systems that run the company without the leadership team becoming the bottleneck. But beyond systems, who you hire is arguably more important. With the wrong people in your team, your systems don’t really matter all that much. If people want to flee from responsibility, they’ll find a way to make it happen. Instead, I’d argue we should be building teams that embrace ownership. It’s hard to screen for someone like that in an interview environment, but once something goes wrong, and something always does, you’ll know whether or not you’ve made the right decision. Which dramatically helps the leadership team to the point where they’re no longer the bottleneck.

The AI-written deluge I get of the "I've long admired your ability to _______. From _____________ to __________ you've consistently ______________" format is very obvious and isn't getting you funded.

“Down But Not Out” is out today! For two years, I asked founders one question: what was happening on the inside, when from the outside nothing seemed to be moving? Yesterday, everyone who answered was in one room, celebrating their own experiences and learnings from their journeys. Teaching us all a thing or two about persevering in entrepreneurial journeys. Thank you again to @anadianant @avinashraghava couldn't have done it without you guys.

keyboard_arrow_left Previous keyboard_arrow_right Next

Love this. Any suggested resources or platforms that are recommended for building your own custom harness ? Anyone else experiment with this and finding success?

Brian Armstrong @brian_armstrong

3 days ago

How to keep AI spend flat while token usage grows exponentially: Not with friction and spend alerts. With better defaults, routing, and caching. Better Defaults (not Usage Caps) – Engineers can choose any model they want, but defaults matter. We’re experimenting with defaulting

451 706 6K 3.7M 6K

@jelanifuel There’s speculation USG will also ban new OSS models from being used Insanely dumb

USG blocking frontier AI models just adds more fuel to China shipping better open source models. And they’ll be cheaper. At that point why would anyone stick to American models? Pretty dumb on their part imo.

Most product team job descriptions I’m seeing will look outdated within a few years. The assumption behind most of them is that depth in one lane is the most valuable thing a person can bring: design, engineering, product, pick one and stay there. The teams I'm watching build something interesting are organized differently. They're hiring people who can move across disciplines without losing judgment in any of them, who understand the business problem well enough to evaluate whether the technology is actually solving it, and who can direct AI doing the specialist work rather than compete with it. I've been calling this profile the "specialized generalist" because I haven't found a better name for it yet. Maybe “M-shaped generalists” is another good term? I’m unsure. But hiring these types of people will have outsized returns in the near future.

nvidia just swept mlperf training 6.0. their blackwell chips are doing the heavy lifting that makes ai models actually work. no cap, the performance gap is getting wild

Congrats to Rocket Lab who just completed a satellite deployment in 16 hours and 42 minutes from the moment the Space Force picked up the phone. For context: it takes 14 hours to fly nonstop from New York to Singapore. Rocket Lab put a satellite in orbit faster than a commercial flight crosses the Pacific. On-demand space access is no longer a concept. It's a product.

I think often about my misprediction of remote work. pre-2019, it seemed like a no-brainer globalism-pilled prediction to say: once remote work technology gets good, it'll dominate, we'll hire more internationally, teams will be more global. you know all the epic industrial agglomeration effects that workers in certain cities get? well, we'll make their AoE range global, and thus scale their effectiveness a couple OOMs. but it seems to be like fully disproven. the main way hybrid work + remote work seem to work is as alternate renumeration for companies with non-competitive pay - with the implicit understanding that you're also maybe just working less. the current model of remote work is thus closer to: "remote work as off-ramp from 40h work-week towards 20h work-week" which is reasonable and cool, but it's still so surprising to me that the defacto state of remote work ended up the sleepy guy option and not the massively agglomerative option.

Here’s a hack that added $30,000+ in MRR for my startup: Your competitors' comment sections are full of people already paying attention to your market. And most teams just ignore them. If someone in your ICP is liking and commenting on competitor posts, they’re probably not random. They’re reading about the problem. They’re comparing options. They’re at least curious. And you can see all of it in public. Here’s the simple version: Step 1) Make a list of your top 3-5 competitors. Add their founders and sales reps too. Step 2) Look at who’s liking and commenting on their posts. Step 3) Pull out the people who actually match your ICP. Step 4) Reach out without making it weird. Don’t just say: “Hey, I saw you commented on [competitor] post.” More like: “Saw you’re looking into [category]. We take a slightly different approach to this. Might be worth a look if you’re comparing options.” Step 5) Keep track of who keeps engaging, then move those people into a proper sales conversation. That’s it. Your competitor is already attracting the exact people you're trying to reach. You just need to pay attention. Ps : gojiberry(.ai) tracks competitor engagement automatically, so check it out !

"10x faster with AI" In practice it's closer to 1.2-2x. You get a prototype in an hour. Then the iteration, the judgment calls, and the cleanup take days.

🎉 Excited to announce the first conversation coming soon. Grant Winns is a 4x exited founder & CEO of Yuna Health, an AI mental health support app. He knows what the founder journey costs you mentally, and he's done it successfully 4 times. 👇🏾 thefounderexhale.com

there's also a new view for logs, we redid the whole thing from the ground up clankercloud.ai

AI is creating a strange new failure mode: People can contribute work they never actually reviewed. That’s not a prompt injection problem. It’s an ownership problem. As code generation gets cheaper, accountability becomes more valuable.

Mitchell Hashimoto @mitchellh

2 weeks ago

Got em. I poison my AGENTS.md (and other things like code comments) all over the place with prompt injections like this to find people who don't review their code and sling it off to another human. Catches folks all the time and then its an instant ban. As I've said, I don't

180 437 7K 768K 2K

When you launch a startup you’re signing up for years of pain for decades of joy, accept that, get through the hard times and keep going.

Watch the full episode with Patrick Collison: dwarkesh.com/p/patrick-coll…