AI & Automation Consulting, Service
Building the future of personal AI ownership.
Open-source agents, local models, automation, and community-driven tools.
Dark matter is the laziest theory I've ever heard of.
Your equations don't predict the observations, so instead of revising the equations, you say: "They will work if we believe in something we cannot see or touch."
God is a better theory than that.
@jun_song Can confirm. I take 20 minutes to garden every day, the rest is building, imagining concepts, testing, iterating, naps when convenient. Even sleep has become an afterthought. Every minute spent now compounds our futures. The acceleration will be insane soon.
@forgebitz Closing the loop isn't hard, but doing it elegantly takes time. You'll start seeing a bunch of people who have been building in private come forward soon with fascinating projects.
@steipete@struggler_rick@ymir_ke I was gonna say, I've been doing this in my own harness too, it's just a good technique. Applying it is not difficult.
Built an open-source, real-time procedural soil & terrain studio using @threejs & @webgl ๐ฟ
GPU-displaced mounds, dry cracks that groove the surface, moss layer, GPU-instanced grass, volumetric clouds & Model import with a moss accumulation.
Code: https://t.co/WZNip6kCzc
Still cookin up. 4 things at once. Harness, Browser Extension, Our Core Agent Loop in Rust, Codex Launcher (it just lets people easily switch between their openai oauth and a local openai compatible endpoint in a gui).
Who else is BUILDING useful local-first open source tools?
Grok Build's usage indicator (buried in the website's settings panel) changed it's look and feel today.
The look was an improvement, but it feels like they're absolutely fucking me on usage all of a sudden, starting today when my weekly reset happened.
70% wkly used in 40 min..
@theslowtell Control, reliability, and auditability (not just trust)(not just auditing the code, every path of work is auditable by scope). Compliance is being handled per client and their needs thus far, eventually the plan is to build as many forms of compliance in as possible.
@art_zucker Not just IP compressed in a modelโ
more like a ship in a bottle.
Few know how it got there.
Even fewer can make it sail.
You're paying for the magic trick. It's entertainment to most.
FABLE 5 CAME BACK NERFED.
We re-ran the July 1st version of Claude Fable 5 on BridgeBench.
The results are brutal:
Debugging: 86.2 โ 25.9
Refactoring: 73.6 โ 38.4
Hallucination: 75.9 โ 61.7
The new guardrails are kicking in on way too many tasks and falling back to Opus 4.8.
This is not the model that got banned.
Anthropic owes everyone an explanation.
Edit: Composer2.5 does NOT eat credits nearly half as bad as I presumed. The change seems to mostly be involving the Grok Build model with 512k context.
..paying the most they allow me to without a special order, and running into this change all of a sudden...
I'm just glad I got my 100$ worth before they switched it up and made it garbage just like every other company has done. I'm now afraid to even switch back to Cursor.
@dannyisr@theo Same! Composer2.5 via Grok has been an insane subsidized value, definitely continuing it for the next 2 months at least. Grok Build EATS credits when not on composer though, for anyone wondering.
@colinsolvely@openclaw UI is one of the hardest parts. Just keep at it. No one can nail it without a good feedback and testing program. I'm learning this the hard way as well with my own unreleased thing.
Today, we're launching the largest open-source dataset of gaming data for computer-use.
500+ hours of gameplay screen recordings + keystrokes/mouse movements, across Valorant, Minecraft, GTA and more.
Link in the comments :)
@markov__ai