Will Hughes

A Practical Look at Regional Production Networks in Modern Manufacturing uklistings.org/A-Practical-Lo…

ORNL researchers are using neutron scattering to map strain caused by residual stress in materials. By comparing how welds perform in different environments, scientists can design materials better suited to withstand the extreme forces of space launch 🚀 bit.ly/4bO0h4h

@AzzyDesignWorks Cleanup day turns into repair week. One bad bearing, zero local stock. UK supply chains are brittle. UAE's redundancy is what I need.

Not where I wanted to be on a cleanup day, especially with several order that need this machine running to fulfill. Crap. Support suggest I might need to source the entire assembly from China, (bad bearing) and attempts to disassemble and DIY have not gone well.

@AzzyDesignWorks Brake cleaner and blind faith while I wait for Chinese parts. Two pillars of UK manufacturing.

Well, its cleaned and back together. Running a lot smoother. Still has the same bearings, but after taking it apart and hosing stuff down with brake cleaner, it got a lot smoother. Lubed up what needed to be, and lets see if it cuts round circles again.

@joshuamschultz Framework-heavy with the classic books-to-action gap. My woodshop shelf has the same bug: 47 manuals, 3 finished projects.

I had one of my agents (Olivia) put together a psych eval purely from the books on my shelf. Not too far off! A few excerpts...

Someone just shipped a 9M parameter LLM that talks like a fish. Trained in 5 minutes on a free GPU. One file — data generation, tokenizer, training loop, inference. Swap the training data and it talks like anything else. This will teach you more about how language models actually work than 10 courses.

I found the GOAT of blue collar builders. @SansScott built goat tracking software with Claude Code and his story is incredible. Scott owns a land clearing company called Hudson Valley Forestry. One of their services is targeted grazing. Some slopes are too steep for machines, and some have high pressure gas lines underneath. So to clear the land, they bring in 60 goats and the goats clear the whole thing. But to scale, he needed a way to track the goats. And his current software said no to his goat module feature request. So he built the GOAT software himself... with no previous experience coding. You need to hear the full story from him. But one of the best parts of the conversation was when I asked him what the biggest impact has been since he started building his own software. "The more I build, the more I can be at home with my kids. And that's better for me." No software company would have built any of this for him. But Scott doesn't need a dev team anymore. He just needs the domain expertise.

@chris_j_paxton Half our UK factories can't get basic PLC programming right, but I'm meant to believe Tesla cracked embodied AI with DoorDash footage? Pull the other one.

Whenever I see stories like this that I know are getting important details wrong, i start how much of the rest of this we can trust The only artifacts about robotics anyone should really, truly trust are public ones, not back room rumors

@zanehengsperger Cemetery to CNC. That's the pipeline. Meanwhile the UK funnels kids into debt and wonders why manufacturing stagnates.

We are hiring operators, apply today: jobs.ashbyhq.com/nox-metals

There is no manufacturing labor shortage in America, there is just a shortage of direction and knowledge of opportunity. There are plenty of young, hard-working, ambitious, talented young men and women in America that should be well connected to the manufacturing labor force. Instead we funnel them into college, load them with debt, and spit them out into jobs they hate. Or they go straight from highschool into something they don't care about. Our last 2 manufacturing hires were so cool. One worked a cemetery digging graves and the other a career carpenter. But now they have bought in our greater mission to Reindustrialize America. With a strong mission, these young adults can learn quickly, gain valuable skills, and haves decades of career in front of them - that is at very little risk to ai. We need a path for young people to know this career path exists. That building with your hands is cool and is fulfilling. And while we work a lot, we are having a ton of fun.

@zanehengsperger AI is absolutely coming for the factory floor. I write this code. UK manufacturers pretending otherwise is why they're ngmi. UAE at least prepares for it.

@miolini Batch size = 1. Instead of stamping out 1000 identical parts, you craft one bespoke unit. I chased this for years in UK factories still using fax. Same principle: one unique company per customer.

@woodtechwill Explain please lot-size-1

We are entering a time when a new company can be created and operated for each individual customer within hours. One customer, one company.

@debreuil Exactly. Wrapping AI in DOM manipulation is like running CNC on punch cards. IL renderers match the actual paradigm.

It feels like there’s going to be an explosion of pet projects people have had on the side burner. There should be a centralized resource to catalog and run these. I think we need a better ’ai browser’ based on IL code and renderers, rather than html.

@DataChaz 1M users per employee is absurd leverage. Guess they actually tested their product instead of hiring 200 PMs to analyze it.

At no other time in history was this possible. Obsidian is a $350M company built by 3 engineers, and their operating system is wildly unconventional: • ~1 million users per employee (7 full-time staff total) • Fully remote with only 1 in-person meetup per year • No scheduled meetings; deep focus is the priority • CEO recruited from their Discord community • Zero VC funding and a hard cap of 12 employees forever • Zero analytics: they actively refuse to track their own DAU What a wonderful time to build.

@StevenGlinert i've found the basilisk fears backup generators. UAE infrastructure handles these theatrics just fine - everything performing to spec.

Apparently you can’t psyop these motherfuckers with roko’s basilisk.

The killer idea: close the loop between actuator design and motion. CMA-ES proposes designs, regression models turn them into real motor physics, trajectory optimization tests them under constraints, and the system learns what works. Result: task-specific actuators with better efficiency, control, and performance than generic designs.

Happy Easter! Taking a few days off this week, so might not have a video, but excited for the SBCC next weekend! sbcc.sdsc.edu/main-page.html

@nurijanian Shop floor rule #1: engineers see the blind spots. Took 7 academic clusters to figure that out?

Your engineers and designers know your blind spots better than you do. Now there’s a workflow for that. PM OS now has a 9th workflow built on John Cutler’s research into what designers and engineers actually said about their best product managers, distilled into 7 behavioral clusters Honestly it's crazy that you can have this and you don't need to have all those elaborate ChatGPT/Claude Projects anymore. New: 10 skills added (205 total)

@robotsdigest +56% is massive. Finally eliminate manual reward shaping and deploy flexible automation without teams of RL engineers babysitting every task variant. UK factories need this yesterday.

The key idea in VLLR is simple but powerful: Robots fail long tasks because rewards are either ❌ too sparse ❌ too handcrafted So they introduce: • semantic subgoal rewards (LLM + VLM) • internal confidence as reward (self-certainty) External structure + internal signals = generalization.

VLLR tackles one of the hardest problems in robotics: long-horizon reward design. → LLMs decompose tasks into subgoals → VLMs estimate progress (for value init) → Policy self-certainty gives dense intrinsic reward Result: up to +56% success without manual reward engineering

@robotsdigest So I can stop debugging sparse rewards at midnight? Good. Decomposition approach makes sense.