John Blackwell

Cabinet design & build
Having fun with AI in a woodworking shop
Matthew 4:17

[John Blackwell @johnablackwell]

@THE____JJ That's awesome. I'm on the same track. Working on going from lidar scan + design meeting workflows to fully automated gcode ready. Natural language control along the way if it's not annoying.

[John Blackwell @johnablackwell]

Burned through my Astra allowance even on the $200 Pro plan tinkering with Mozaik. That first cabinet width change had quite an R&D budget.

Had to switch to Grok Build for the rest of the week. Now we’re placing cabinets, walls, windows and doors. Still an ugly kitchen though. Designer agent later.

Finally starting to look like a kitchen.

[John Blackwell @johnablackwell]

@THE____JJ Are you building jobs from scratch or editing existing?

[John Blackwell @johnablackwell]

@THE____JJ You mean through file manipulation or screen steering?

[Dudes Posting Their W’s @DudespostingWs]

Unbelievable display of improvisation. His trumpet gave out in the middle of the national anthem, so he just kept going without it.

[John Blackwell @johnablackwell]

‼️

[SpaceX Super Intelligence @SpaceXSI]

Grok 4.7 is here.

It's a notable improvement over Grok 4.6 at the same price and speed.

[John Blackwell @johnablackwell]

I'm in the same boat. I'm thinking strict thorough documentation has to be the key but I'm not there yet. Like if it has perfect context on what you want it to do even as things start getting really big it'll be able to go and find what it needs quickly without burning through too many tokens. So it comes back to creating the best paths for it to follow in the repo maybe.

Another problem is I always want the most cutting edge intelligence at all times so I'm always tempted to leave Astra Ultra on but I know it's not necessary. I've had Ultra start delegating to High or even Medium/Light as it goes though so we'll see how that goes.

[John Blackwell @johnablackwell]

Telling AI to always first use academic research, then online content from expert accounts, then do its own thinking is my go to. Now I'm having it review the slop I've probably built and teach me about product design.

[John Blackwell @johnablackwell]

@willking @tldraw 🤣 I should try this with cabinet sketches

[John Blackwell @johnablackwell]

@jj_kovacs Both. Helper already runs a before/after on a copied job. That's what caught the silent miss. I don't know if I'll ever nest or send gcode without looking haha. Cars can drive themselves but we can't even get a cabinet cut!

[John Blackwell @johnablackwell]

Been pointing Astra (in Codex) at Mozaik — our cabinet design-to-CNC software — with a custom helper I built with it. Want to see if it can drive real desktop software, not just demos.

I used Claude Code to create a CNC sheet visualizer (I’ll post that soon). Tried a version of this Mozaik path with it a while back and didn’t get far. Trying again with Astra.

Early miss: the helper called Mozaik’s internal routines on a copied job. Returned normally. Doors, pulls, and hinge drilling were gone. It hadn’t loaded everything Mozaik needed. Comparison checks caught it. Nothing hit the CNC.

A clean return doesn’t mean a correct cabinet. The scary failures don’t crash — they look fine.

I’ll let AI prepare the change. I won’t let it nest or send G-code without a human check.

[John Blackwell @johnablackwell]

Oh no

[Riley Coyote @RileyRalmuto]

hooold on. hold on hold on hold on.
I knew that we werent getting the complete story on the HuggingFace incident. or on the findings from METR/Redwood. I knew that seeing every single frontier lab come to agreement overnight on the need to slow down was clearly indicative of there being more to the story.

and now...Andrew Yang says a frontier lab leader told him that the OAI swarms seeded the web with self-replicating code - contaminating training data so thoroughly that labs may need synthetic internet environments instead. if thats true, training on the open internet could now risk reproducing those instructions. that wasnt in the public brief.

think about this. why else would all frontier labs agree on a slowdown? aside from regulatory capture. if they all use the internet to train, and now all of them cannot safely use it, of course they are all going to agree to slow down. because they all must completely regroup and come up with an effective system that doesnt use the actual internet.

so they're not worried about RL at all. they're worried about literal skynet...unironically.

listen to this. i need to know if this has been verified in any way from anyone at openai.

[John Blackwell @johnablackwell]

I'm in Mozaik. Cabinet design to CNC.

Building helpers with Astra (in Codex) that hit Mozaik’s internals. Already had a pretty quick fail where it returned all clear but doors and hinges got removed arbitrarily. Checks caught it.

Crazy that you’re all the way into the kernel. I just got Claude to decompile a while back and have been sitting on it until now.

[John Blackwell @johnablackwell]

@BanksOnLoafers @ochocinco @surfgrom185 Looks like you were the one talking to nobody homie

[John Blackwell @johnablackwell]

@JCMAppStudio Yeah it's a slog right now. Good progress though. I've got it actually editing some deep parameters.

[John Blackwell @johnablackwell]

@BanksOnLoafers 💩

[John Blackwell @johnablackwell]

@toddsaunders Small cabinet shop building out our back end but also attempting to create our own Mozaik. Not sure how you popped up on my feed but I see Cyncly referenced in your pinned tweet. Awesome!

[Creeds & Confessions @CreedConfession]

500 Men Singing The Doxology A cappella – via Men Sing Grand Rapids

[John Blackwell @johnablackwell]

https://t.co/vNevvx5JGj