Chris Paxton

This isn't AI "eating" the economy, its that only one part of the economy is still growing/doing well.

Seems bad tbh

We got a good couple of brutal wars out of it, though, while we figured it out

I mean he's not wrong but this would be... bad. This is a bad thing. People should not be doing this.

What the hell are we doing here

Thanks!

Gotta practice my whistling and egg-laying to remain hireable

Yeah

Why Blue States Can’t Have Nice Things In the tangle of outsourcing, no one is actually in charge.

Oh this is from the very excellent Persuasion newsletter: open.substack.com/pub/persuasi...

This gets to the core of it: "What’s missing is the basic expectation that public employees are directly responsible for solving public problems, and the role we as citizens play in creating a culture that makes good governance possible."

You rarely expect public servants to, you know, help

We are going through this in thoroughly democratic Pittsburgh where roadwork and maintenance stretches out indefinitely and nothing is ever fixed; random closures every hundred meters, sidewalks shut down, the works. Total failure of state capacity.

Whole body controller from my team. Lean on it, push it, robot doesn't fall. Important to watch the feet: it smartly changes its posture as Jonah pushes down on it.

At least there's no moat deep enough to stop some random Chinese team with like $50 of bootleg gpus lol, Microsoft and Amazon can't figure it out and nvidia still has to fine tune qwen so...

Yeah that tracks

Genuinely what the fuck, how?

A side-by-side bar chart compares the performance of three models—XBai o4 (blue), OpenAI-o3-mini (gray), and Claude Opus 4 (yellow)—across two operational modes: Medium Mode and Low Mode. Each mode has results on three benchmarks: AIME24, AIME25, and LiveCodeBench v5. Medium Mode: • AIME24: XBai o4 (85.4), o3-mini (79.6), Claude Opus 4 (75.7) • AIME25: XBai o4 (77.6), o3-mini (74.8), Claude Opus 4 (75.5) • LiveCodeBench v5: XBai o4 (67.0), o3-mini (66.3), Claude Opus 4 (61.3) Low Mode: • AIME24: XBai o4 (82.4), o3-mini (60.0), Claude Opus 4 (75.7) • AIME25: XBai o4 (74.8), o3-mini (48.3), Claude Opus 4 (75.5) • LiveCodeBench v5: XBai o4 (66.6), o3-mini (62.0), Claude Opus 4 (61.3) XBai o4 leads in nearly every category, with particularly strong performance on AIME24 in both modes. Claude Opus 4 closely trails o3-mini in some Medium Mode results but outperforms o3-mini in Low Mode for AIME25.

XBai-o4: a new supermodel

* Open weights, apache 2
* 32B
* beats o3-mini
* for TTC they train an extra head as a reward model to do binary classification

hf: huggingface.co/MetaStoneTec...
paper: arxiv.org/abs/2507.01951

Thank you!

It depends on what you mean? If you took asimov or someone forward to ten years from now, they'd certainly think we had achieved it, yes. I think a superintelligence doesn't need to be superior in all dimensions, just many.

An airplane is not a bird, after all, and though a bird could never dream of the speed and altitude of an airplanes flight, there's plenty they can do that planes can't.

Likewise, human intelligence may well have comparative advantage even in an age of superintelligence.

I confess I read this one a while ago, and it’s very much stuck with me. I keep coming back to it. It’s one of those bits that, once you’ve spent a lot of time around little kids, just feels too real, so it sticks with you and hollows you out a bit inside.

substack.com/home/post/p-...

100% agree

The most important skill right now is to be adaptable and open to the fluid and the abstract (specifically, the uncanny).

For most of human history, there was no (appreciable) technological change within the course of an average lifespan. We live in uniquely dynamic times, and therefore the only way to stay afloat in my opinion is to be willing to change

The "defeat" part is still in question but to me this makes it kind of the series of our era

This is basically Robert Jackson Bennett's new book series (most recent book, a drop of corruption) for what it's worth

One thing about humanoid robots i think is under appreciated, is that you dont have to make some super crazy training set for some specialized arm on wheels. It can learn from humans performing actions, so this is incredibly useful

The biggest driver of robotics innovation over the last couple of years hasn't been anything clever. It's just been people making cheap robots easier to use. And that's a good thing; we have all the tools we need for a robotics revolution, we just need to actually get people out there using robots.

One thought that's stuck with me is how much of recent robotics research -- so many startups and papers -- have just acknowledged that we should make it really, really easy to use robots and to collect data. I wrote about it here: open.substack.com/pub/itcanthi...

It seems that many AI companies and research labs — including ByteDance — are exploring Diffusion-based LLMs.

Seed Diffusion: A large scale language model based on discrete-state diffusion, specializing in code generation, achieves an inference speed of 2,146 token/s, a 5.4x improvement over

seriously lol

*Reading all the comments under this post*

I thought you guys wanted the AI that did the laundry