Strong agreement on the thesis. The piece is most useful for naming what's bankrupt about prompt chains. Where it stops short is what the verification checkpoints should actually verify.
One way to slice it: there are three kinds of underspecification an agent has to close.
Intent: what the user wanted (JWT vs cookies, should free users see this feature). Verification can't close this and probably shouldn't try.
Structural e.g. null, types, exhaustiveness, ownership. Sound static analysis closes this by construction.
Domain e.g. auth on every route, error propagation, contract stability. A domain-shaped apparatus closes this because it knows what kind of program is being built.
Babysitter, auditor, prayer is the right taxonomy of bad options. The fourth option is making the LLM a component inside an apparatus that handles structural and domain statically, and leaving the human on intent.
We just released new capabilities in Skip to make reactive services scale easily and efficiently across machines. Skip’s core is a native reactive computation graph that minimizes unnecessary recomputation—great for real-time apps—but now you can also scale horizontally using a leader/follower architecture.
We’ve open-sourced an example service (yes, it's called hackernews) you can try locally or on Kubernetes. It supports seamless scaling via kubectl scale, with zero downtime and no changes to your backend logic.
Let's not forget about Tipeee, which might be the only platform actually not in debt (they merely raised 150 000€ 3 years ago, and officially declared benefits last year).
As a content creator, what I need is the ability to create multiple projects under one account. I want people to donate to separately to each individual piece of content I create.
Ideally, a payment platform should offer an API for creating projects programmatically and retrieving information on the funding.
It's on the wiki of a still-in-progress platform that actually plans a novel and potentially better model than all the other things, at least in terms of funding public goods without paywalls (many of these sites are paywall-focused, freemium projects with some public, some exclusive publications).
One way to slice it: there are three kinds of underspecification an agent has to close.
Intent: what the user wanted (JWT vs cookies, should free users see this feature). Verification can't close this and probably shouldn't try.
Structural e.g. null, types, exhaustiveness, ownership. Sound static analysis closes this by construction.
Domain e.g. auth on every route, error propagation, contract stability. A domain-shaped apparatus closes this because it knows what kind of program is being built.
Babysitter, auditor, prayer is the right taxonomy of bad options. The fourth option is making the LLM a component inside an apparatus that handles structural and domain statically, and leaving the human on intent.
reply