athrowaway3z

Without actual code solving an actual problem discussing this is all a bit wishy washy, but I think you're mixing 'levels'.

The sans-io design here is about building reusable parts that any runtime or thread can drive and incorporate into the colored interface they expose.

Function coloring is about (the annoyance when) composing interfaces designed around those different runtimes/thread that might use the same sans-io library internally.

Yes async compiles down to a statemachine, but we're talking about the elegance of the language syntax/composition. Most languages require _a lot_ of ritual (or overhead) if you want to use the async-statemachine as a pure statemachines and sidestep the build in runtime-side-effect channels.

Having used herdr since its release, the recent 'Emacs is 50' made me decide to go back to it and just vibecode herdr features i use into an emacs package.

It works great so far. I can just ask it to update its layout whenever i think something better is possible. I wont be going back, but the one thing that might change that is a mobile app.

Its actually not settled yet wrt copyright.

However, there is a more straightforward solution. The allows general use inside a company. The license blocks the "Model-as-a-Service".

So instead, any GPU holder can provide a 'Infrastructure-as-a-service' and have clients provide the model, and then tell the clients that as an optimization they can also provide just the hash and they'll load it from their memory.

I think its good that this license pushes the ecosystem towards such a generic-utility framework, but as a moat to protect a business its banking on most companies being lazy and wasteful.

That pays off more than people like to admit.

This comment inspired me to get back into emacs. Had claude rebuild my pi+herdr setup with pillish and a custom emacs package.

This is pretty cool, but at the same time i'm horrified that somebody might walk away thinking you need to deny these requests to protect against XSS or SQL injection.

Maybe i want to sell coffee mugs with `<img src=x onerror=alert(document.domain)>` on them.

Parse, dont validate. (And never interpret what shouldn't be)

I could have been more precise, but it's the same ideas underpinning HTTPS and git - long before blockchain became a hype.

Did you instead want to build some central hub that grounds all train traffic if unreachable, or create a bunch of semaphores that anything can overwrite by accident or on purpose?

I'm genuinely curious here.

I can understand being intimidated by concepts outside your wheelhouse, but if you have a simpler solution you'd feel safe with i'm all ears.

I'm not saying this is all there is too it, but I am saying all the other design choices like recovery modes, operator experience, etc - should build around cryptography between tracks and trains.

We can call them cryptographic semaphores if that makes your brain turn on again.

On a related note from that tweet

> Noteworthy: ironically, OpenAI and Anthropic's models refused to do this work. Open weight models got the job done. I'm worried the labs' stance might backfire and get in the way of helping developers build correct and secure systems.

I was working on a network protocol and it correctly found a DoS attack. Once found - Claude refused anything and everything to do with it. Even asking what it was reading locked immediately. I had not written any of it.

Sol 6.1 only denied me once.

Having a multi-model harness was always the better choice, but this really drove home the point to not get too dependent on any single provider.

Who wants to design a basic cryptographic p2p lock between trains that limit their speed 30km/h if they're not able to get proof they're alone on a track; when the alternative is to debate endlessly about whom to blame, and which billion-dollar conglomerate has the technical prowess to sit on the committee to debate the specs for a much worse system.

Ok i should not be trying to wrap my head around this, this late at night - but i got to be frank and tell you this is way too loud and indecipherable. I need a two sentence explanation given that i work with `pi` and build my own tools.

Do i have it right that its some prompts injection for reflection and a composition scheme for the prompt/context where each element is version controlled?

If so, my experience with all the SOTA models is that they're pretty bad at finding the 'clean' solution to self improvements; eg they still always add text instead of taking it away, and somebody needs to remember to tell them (eg AGENTS.md) to delete stuff.

I suppose there is a real demand for a harness that does get better over time for average users, i'm just a bit skeptical if that actually works out over a longer time.

Might check it out later.

Do you stand by your blaze attitude?

This reads like you want to gate-keep people first exploring and discovering the absurdity of the surveillance state we live in.

In my opinion, Claude "gaslighting" is a signal the other way - If Claude thinks they've discovered something, that means the llm-compression of most digitalized human knowledge still finds SmartThing a disturbing piece of tech that user do not want.

An unexploited PHP server means "there exists a configuration that is not exploitable".

Which is how hundreds of previous exploits have had people publish information on what to temporarily tweak/disable/block, such that the exploit wouldn't work.

What is novel about this case, and for which i do not know the explanation, is that nobody in-the-know was willing to confirm them before the CVE dropped.