@mattorb

Stay curious. Fan of science and design.

 Dallas / Fort Worth, TX
mattorb.com
Joined May 2007
  • Tweets4.6K
  • Following659
  • Followers659
183Media
🤯 Experimenting with Jev on Cloudflare Workers AI for quickly&cheaply assessing PR's and Opus 5.5 for video generation. Try it live at PRedict.mattorb.com/ It makes me wonder how TypeSafe AI does evals for a model which claims to classify and score literally everything!
1
1
177
Added clef and clef-flash via Cloudflare. For reference the Clef model card (huggingface.co/Cloudflare/cl…) has a few evals they score on business workflows clef vs clef-flash vs jev. Typesafe Evals here: evals.typesafe.ai/
1
86
I'm winding down a very broad and deep tour of building lots of different things with AI, for fun. It involved everything from creating a distributed AI job/harness scheduling system to execute repeatable job chains with unique strategies in vendor harnesses, to an in-loop AI
2
1
114
oops, still tuning syndication.... links here: mattorb.com/thoughts/winding…

I'm winding down a very broad and deep tour of building lots of different things with AI,… - mattorb

I'm winding down a very broad and deep tour of building lots of different things with AI, for fun. It involved everything from creating a distributed AI job/harness scheduling…

mattorb.com
50
It has been a blast!
1
20
Probably both the most visible and the thing I have heard the most positive feedback on is my personal site revamp, so I wrote up (with some help, of course!) a recap on all the little customizations and niceties I pulled together there: mattorb.com/customizing-matt… Cheers!

Customizing mattorb.com to my taste with AI - mattorb

After moving mattorb.com to Astro, I used an AI coding agent to build the weird, specific features no blog template would have given me.

mattorb.com
54
Doing a little experiment with having a bot research agent that checks LLM generated prose phrasing of claims against sources.
Diagram titled Evidence gates: a research pass produces an article and a matching dossier, which pass through three build-blocking mechanical gates, then an adversarial verify step and a blinded judge panel, before a deploy guard that refuses on a moved hash; unsupported claims loop back to the sentence or the record.

ALT Diagram titled Evidence gates: a research pass produces an article and a matching dossier, which pass through three build-blocking mechanical gates, then an adversarial verify step and a blinded judge panel, before a deploy guard that refuses on a moved hash; unsupported claims loop back to the sentence or the record.

1
122
Got a 3D printer recently (P2S)... Learned a bare minimum amount of CAD and then discovered gpt-5.6 can do quite a bit without it!
A chat assistant revising a 3D-printable model on request: it reports moving a Touch ID opening 31.5 mm farther right to the upper-right of the removable cap, cutting through the top shell only with the underside left intact, and verifying the STL and 3MF stay watertight as two printable parts, with download links beside a 3D viewer showing the part on a print bed.

ALT A chat assistant revising a 3D-printable model on request: it reports moving a Touch ID opening 31.5 mm farther right to the upper-right of the removable cap, cutting through the top shell only with the underside left intact, and verifying the STL and 3MF stay watertight as two printable parts, with download links beside a 3D viewer showing the part on a print bed.

1
1
127
and revise....
1
51
TIL: ASD-STE100 ftw!
A terminal AI coding session asked to use ASD-STE100 Simplified Technical English, contrasting its default conversational style with controlled-language rules: one term per concept, one meaning per word, short sentences, active voice, explicit references, no idioms or contractions, and repeated terms instead of synonyms.

ALT A terminal AI coding session asked to use ASD-STE100 Simplified Technical English, contrasting its default conversational style with controlled-language rules: one term per concept, one meaning per word, short sentences, active voice, explicit references, no idioms or contractions, and repeated terms instead of synonyms.

95
Also in 2026: new tools enable new possibilities.
A Claude Code session where the user asks for help sorting a 650-item read-later queue, loads a Keep skill, then asks to archive all iOS-related items so they leave the active queue but stay searchable.

ALT A Claude Code session where the user asks for help sorting a 650-item read-later queue, loads a Keep skill, then asks to archive all iOS-related items so they leave the active queue but stay searchable.

78
Prototype of adding HTML plan file support via in-loop browser launch of 'rep' tool. It won't work cleanly over ssh/tmux/etc., but great for local HTML plan revisions. Uses same keyboard shortcuts as the TUI, for quick plan navigation too.
78
2026 collateral damage: I dropped fish as my preferred shell this year because generating zsh scripts with LLMs is just so good that for me, the benefits of using fish no longer justify the weight of the additional dependency in my default machine setup.
1
72
related: also attempting to not use Homebrew at all on personal Mac, in favor of pinned project-level dependencies, and mise installing binaries from GitHub releases via their provider.
33
"One of the challenges we will face in the next 2 years is the economics of tokens. The marginal cost of productivity improvement has to match the marginal cost of the token. That's a management discipline. [...]
1
40
The business has to benefit from it." — SN (overcast.fm/+m_rorM9M4/27:08)
A brass balance scale with glowing computational tokens on one side and rising glass blocks representing productivity on the other.

ALT A brass balance scale with glowing computational tokens on one side and rising glass blocks representing productivity on the other.

41
What's missing?
2
98
Well well.... Hello there. 🤣
95
Load more