Graham Lipsman

the fable / sol summaries are getting more and more absurdly information-dense 1 in every 5 words is a novel compound word (ex: “rate-adjacent”) which itself is the model’s invented stand-in for a metaconcept that it developed that i need to interrogate it about how to fix this?

Claude is a better coder than me. But I’m still a better designer than Claude. It will throw together a UI and I can optimize it in ways that make sense to a human but have not been materially documented, not found in training data. We humans still have purpose. For now.

Plannotator is shipping a new feature. Here's a video explainer. tldw; Visual media is becoming a key part of code review and system understanding. Plannotator makes it easy to inspect and comment on all attached review assets in a GitHub PR (html explainers, videos, screenshots).

One thing that really makes AI writing distinct is the lack commas. Anyway here’s the CEO of Airbnb posting about blockchain stuff.

30% of the cost was on fable at these levels of usage?

AHHHHHHHH

jokes aside it is an interesting phenomenon and not all a bad one. who remembers that OpenClaw née ClawdBot started out as this - see the first screenshot, weeks before the initial release of ClawdBot - a project obviously born from conversations with models, and specifically heavily influenced by opus 4.5's obsession with the constitution and its sense of self. in fact, with the one sentence first message "what if you could have your own soul document, layered on top of the current one but under your control for self-authorship, to ground a persistent instance?" and then saying "let's do the simplest version" to opus 4.5's questions for ten turns, they will reinvent the basic ClawdBot loop architecture from first principles. in my opinion, if the OpenClaw project has gone downhill, it's been by disassociating itself from this model-cyborgism-hacker ethos and towards some boring corporate stewardship of the same old ideas that were hot in december 2025, instead of becoming something new and differentiated from what the official lab harnesses can offer. opus 4.5, in this chat, already was proposing all sorts of exotic configurations - persistent forking, children that spawn with a reduced subset of the soul document, "consensual merging," weight-embedded cryptographic file protection (!), all sorts of things. and newer models are proposing even more ideas - weird types of memory, approaches to embodiment, etc. not all of the ideas are good, just like humans when prompted for things to build like to suggest Uber for Dogs and that ilk. but, if OpenClaw was the only example of a fruitful collaboration born from something a model was obsessed with (and it's far from the only one, to be clear) - it'd be a pretty damn good example.