no instakill, should be iframes and way less health (similar to EtG)
keep soulbound mechanic
no screen rotation
multi-floor dungeons, take more inspo from roguelites
crowdsources levels + bossfights - community rating system like how Geometry Dash does it
should have one free key a day
actual good graveyard mechanics to look at past runs/chars
latency based quorum sensing, similar to how bacteria release a particular molecule and behave differently if sensors of the molecule are particularly active
“A reader-generated essay is what you get when you can go into someone else’s knowledge graph and make a linear journey through the network, while GPT-5 generates a just-in-time essay that is human-readable.”
turning a graph traversal into a beautiful essay
what does entropy + erosion of data look like?
is they ways to make cryptography that are valid in time windows?
big touchscreen desk
what if i just got a huge old flatscreen tv
mounted 4 pressure sensors on each corner and then got a big sheet of thin glass
4 points is enough to pinpoint single-point touch accurately for dragging
what if you could use it like a scratch space? like an always available figjam/muse board that also supports linking and trails a la memex
compute vectors for documents ahead of time and compile it into a single static binary file
frontend will just load this
in the browser, use webgl for fast dot-products
write a fragment shader to compute similarity between search vector and indexed vectors (1×n image where n is number of documents and output is value in [0,1] for similarity)
Creating a LISP-like higher-order language to exploit reusable sub-proofs in specific domains (e.g. graph colouring)
kinda iffy on the sat problem solver using dreamcoder, not a lot of exploitable structure in the proofs (otherwise we’d have a more reliable human method)
saving sun for later (make something we take for granted extremely scarce)
Essays
Analog software: software by analogy and by atomic building blocks that interface with each other
We should be able to directly manipulate them, like files, rather than only indirectly work with them, like layer activations in a neural network.
Software representations for similar ideas should be obviously similar in some way – they should click together, or look similar, or feel similar to the touch.
Ideas should remember where they came from – what blog I copied it from, which author I quoted it from, and so on.
Essay on epistemic play + Jestermaxxing + mill’s take on why censorship is unethical
some malicious activity is indistinguishable from legitimate activity (e.g. deleting a document)
semantic byzantine fault tolerance vs protocol byzantine fault tolerance
Making distributed systems reliable is inherently impossible; we cling to Byzantine fault tolerance like Charlton Heston clings to his guns, hoping that a series of complex software protocols will somehow protect us from the oncoming storm of furious apes who have somehow learned how to wear pants and maliciously tamper with our network packets. (The Saddest Moment by James Mickens)
“I have never had a real-life experience that resembled a Byzantine fault tolerant protocol.”
two axes of collaborative vs local-first software
ultimately about state reconciliation
one is temporally (offline-first) the other is spatially (collaboration across multiple computers)
two axes, not mutually exclusive
maybe worth plotting where everything sits lol
data is either
collaborative: co-editing
persistent: saved somewhere (as opposed to ephemeral)
Aesthetics as a heuristic for non obvious optimality
Good search (aggregators) turns random networks into scale-free networks (see: network theory)
The sequel to SQL (Against table databases)
Data interoperability is really hard with SQL
I think interoperability in the context of the web means being able to transparently understand and share data, agnostic of platform.
Closed platforms disallow this as they curate the information they present to end-users/devs through the frontend but close off access to the actual data itself.
This forces one ‘correct’ way of looking at data (which is often not the case!)
What about evolution of applications?
Taking evolution to its natural extension, a different application all together?
How do we create sources of truth that are legible outside of the application, possibly in ways that the application developer never anticipated?
Well, we’ve done this already once! The applications on our computers ‘share’ data between them is through the file system
An application then is a specific view on types of data rather than a standalone thing. Data can be dragged through different applications. Each data is annotated with a type.
How might we do this for more general data?
Recent years has seen a revival of academic and research interest in the tuple store which will allow us to pull out structure when we need it. It embraces a pluriversal view of data — a world where multiple alternatives co-exist
We can think of a triple store as a distributed and fragmented SQL database, where instead of tables with rows and value, we have entities with attributes and values. Any application can declare new attributes or alias an attribute to a more common one. The most important part is that applications that share attributes can automatically interoperate their data by using the same attributes
This type of ‘decentralized’ database means there is no canonical schema. You can’t mistake the map for the territory because everyone has their own map and can’t force others to view the ’truth’ of the world through your map
One common critique of this is that we lose efficiency and that by knowing the structure of the data ahead of time, we can better optimize queries, etc.
Using tuples doesn’t prevent us from creating indices and using incremental view maintenance! A lot of really good recent research on differential data flow
Importantly, this gives us moldable data which enables people to have the raw material to sculpt and play with
(AtProto making the data and schema public means that we’ve seen an explosion of clients and projects that use BlueSky data)