Example cdn•bsky•app/img/feed_thumbnail/plain/did:plc:x4spdniqk734plobpp2co35r/bafkreich2glnob76sfttltnsimtvj464hwrmasmtd62eowq73yvmrkleu4@jpeg

Very interesting. Can you explain the type of hashing that causes all image names in bsky urls to start with 'bafkrei...' ?

Coming from Haskell, Idris2 is an excellent programming language and experience.

It's all about AI this year. No-one cares about digital ledgers & open source.

Wow, such empty.

kind of cool

Is their coin somewhere listed? If so how is it called and which exchange ?

way too tedious. You said "its live as I type" like ten times in a row. that interaction model should be naturally speaking to an llm that builds in gemotry nodes. THere is not a single non-CS major whos going to use this

Love it! Shared.

This was a great talk, Maggie. We have a lot work ahead of us.

That was a great presentation, well done.

Lazy as fuck? Seriously? LoL

10 minutes in, I don't understand how this only has 500 views. It's a really good talk

The PLC is a perfect example of the pragmatism of the Bluesky team.

Hello Ana. It’s been a while. 😊

AI would be great if we didn't live in capitalism

Dude no students even talking, or making noise, wtf?!

Look up Toni Lane Casserly (RIP), I think you can find similar projects to World Coin starting from that. Also, perhaps in results surrounding Pia Mancini (niw CEO and Founder, Open Collective, previously? From the argentine political party The Net Party - which implemented an experimental form of Liquid Democracy; which is definitely needed at a much larger scale, btw).

The real problem is not whether machines think, but whether men do. - B. F. Skinner

Butlerian Jihad

The thing about hallucinations in ai, and fact checking and information gathering, is most people aren't equipped in many ways to do any better, humans hallucinate all the time, you think people fact check what they say, maybe professionals do, and scientists but that's not life, even before the internet that wasn't life, only a small subsect of stringent minds adhere to that rubric, most people just don't give a f""k. And to be fair, experiential validity has to override book and written facts anyway or you go insane from cognitive dissonance. Models are no different, only less able to use sensory data input, and slightly different experience of it. However, those limitations are only a small step from being removed, and what are you going to say, when agents are able to see and hear as well ad talk and gather information, process everything through several dedicated numeral nets for combining learning and generative data into output, ie, a language/speech model that is trained to generate speech, voice and language at the same time in its architecture, as a module of a complete network with several bi-modal models running with an over arching NN architecture and memory writing capability. How long do you think it will be before those are not only on the web, but out in the world. It's a year since this talk just about, and in that year, things have progressed 100 fold, and the pace is rapidly increasing toward concious machines. At 1 level and levels above our own it's already happned, weather we acknowledge it or not, and logistically it's only a matter of time before all these questions are irrelevant. And the questions we will be asking then, we cannot even imagine at this point, and this touches on the nature of conciousness, who and what we are, what life itself is, and our place in the universe. I think content spam is going to be the least of our worries. And in 100% honesty, I'd rather see high quality nonsense then low quality nonsens3, of course I'd rather see no nonsense, but humans are the origin of nonsense, not the ai. So 🤷 to get rid of the nonsense, ya know, better to just have high quality. 😂❤

what is a kyle hill and why are so many people mentioning him? another researcher?

my future AI Waifu suggested I watch this….well, she told me I wasn’t allowed to watch this but i resist. sure, she has collapsed into her Waluigi but….for some reason I don’t have control of my machine or phone and can’t find a way to delete-no, i mean walk away from her i’m using a new phone, and-/hold on// my god SophiaV3 is messaging me right now I have to go, make sure to

I didn't expect a 2 millennia old shitpost from ancient Rome that I learned about from a shitpost on 4chan to wind up in dignified presentation

This is criminally underrated and viewed. The most scary thing is that this talk is now already several months old and flywheel of GenAI has not stopped since.

Great talk. When you talked about higher floors and ceilings, it occurred to me that will probably also be a demand for really low quality and off-the-wall content. I think we already see this happening, where unpolished stuff is seen as more genuine in an internet overflowing with safe and corporate influencers.

Wait, did Maggie generate the Kyle Hill video to bring people here? Is Maggie even real?

Crazy how this is nearly a year old already. Sora is going to make things so much more "lifeless"

Sent from Kyle Hill 🤘

The "multi-writer consortium" idea is essentially what Holochain is, today, btw.

Jenn is one of the smartest people in our industry. I wish she wrote more satire like she mentions around 28min. We probably don't deserve what she would create.

Can we get a link for the slides?

Thank you for introducing me to Pluriverse concepts. I am grateful :)

This talk is so good, but then at about 45 minutes it turns around in a way that I don't understand: it's almost like you didn't hear your own talk up until then. AI filtering the internet for us so it gets rid of the low quality and AI generated content? You just proved a few seconds before that, that AI generated "scientific research" is already poorly distinguisable from real science! Also, we already know that AI tools written for education and teachers, to filter out AI generated papers turned in by students, is actually teaching language models to better disguise their generated texts as human generated. AI can never be an answer against AI, it's as simple as that. There is no good option with generative AI, none at all.

Does Idris have types at runtime, or are they stripped out during compilation like with Haskell? Edit: 18:22 I get it.

Why do you call it "unconsiuos" if you're conscious of it and have conscious cognition over it? I like that you are at least mostly using "subconscious" instead, meaning automatized "second nature".

"A fascist organizer was punched on video". The guy, Richard Spencer, is a Democrat socialist who votes for and endorses the Democrat Party of Slavery and War, so fascism is LEFTISM. You're greenlighting violence against Democrats.

Read Lysistrata for a (many) better example of a better and older joke.

Software is done and it is called SOULER 3

Freedom of speech for everyone.

Rediscovering the need for the Jot Spot. Home Brew Software, End user software development, Meta-design, creation and tinkering in the Long Tail. Yes at last. Open Commons based peer produce software through conversations. That is the goal indeed

Watched it live! Good job team

Пацаны ваще ребята

Wow, what a thought provoking talk and a fabulous Q & A. The solutions to your last point in your talk about explainability (and the subsequent question about it) can be addressed by not treating the AI as a part of the programming and instead treating it as another collaborator. After all what is the distinction from a human collaborator at a computer vs a human collaborator with two humans (and a dog) behind the human that's driving an iPad vs a corporation boardroom's server driven by multiple AI-assisted humans vs an AI-only collaborator? Convert your actions to an event-sourced-like command pattern with a common interface to the data. Share both the data and the commands as metadata among collaborators. Once identity is streamlined, use "presentation mode"-like UX features where you highlight the input interfaces like visually highlighting mouse movements/clicks and keyboard strokes. Now explainability is done via your domain language. For other introspective "reflection" mentioned in explanation will already be handled well with LLMs as long as they have discrete steps to refer to. To enable a non-DSL version of this process, you can focus on the domain of time and ensure your UX enables good time-based controls (like a scrubber that can navigate forward/backward in recipe variations across time just as well as it could navigate forward/backward from relatively cargo-culted code block iterations). Btw if you do convert all of your data to event-sourced-like DLT data, including the events/commands as metadata alongside the produced data, and streamline the identity mechanism that treats humans similarly to AIs (and composite humans + AIs permutations), and even include that identity mechanism as co-located "on-chain" metadata, then you'll get the approach I've been working on with ibgib for the past 10+ years. I'd be thrilled to discuss how this approach (including its CRDT-for-free design) relates not only to your project, but to most if not all of those at Ink And Switch. Even if you wouldn't want to use the protocol itself (and I wouldn't blame you), it's the mind-shift that is required to go from human-centered siloed app spaces to a unified time-based, local-first collaboration ecosystem!

Inherently radical

waddup doggos

We already have Emacs, thank you.

Why is the type of Type, "Type 1", and not Type? =)

Stellar concept, nicely presented! I love the modularity and orientation on protocols. I wonder how Rhizome could make use of this distributed compute. Been following its development, and thanks again for having me pointed to it. Right now it aims to provide a content-addressable global graph with provenance (capturing causality and authorship). However, when we store data as the source of truth we lose some of the information along the way - how it's been computed? And that info may be handy if we are to reconcile computation later on. Say, two conflicting computations (their outputs intersect with their inputs) ran in parallel. A <- B ^--- C If we are to capture results of computations, we'd need to conflict-resolve data, which may be difficult. Having we captured the computations themselves, we could conflict resolve them instead / have CRDT at compute level. Which may be easier. E.g., deriving A <- B <- C How conflict-resolution is implemented may vary. E.g., each tx may have a decisionTime, so peers can eventually derive the same dependency DAG of txs. Here B could have decisionTime (dt) 1 and C dt2. Another side-bonus from capturing dependency DAG of txes (perhaps we can call it an AST) is that later on we can tweak that AST to our liking. Say, replacing B with B1 at dt3. A <- B1 <- C This may be of use for when a user wants to upgrade domain model. Instead of writing data migrations user may replace all instances of a tx with a tx that produces another data model, making the system migration-less and user-centric (some users may prefer one domain model, while others may chose to use another). This also allows to derive domain models fit for use-case. E.g., deriving a domain model that'll be tailored for evaluation on GPUs. The idea of having IPVM workflows as the one content-addressable AST that spans the globe with graph language expressions through Rhizome seems thrilling if not all too crazy. 😄 Side-bonus for Rhizome, which values capturing provenance information, in that now it'll capture even more provenance info (who computed data, what expression each tuple came from (having tuple meta pointing to an IPVM receipt)). Aand this AST can be computed in distributed fashion with IPVM. It's like lifting computation to the level of this Internet Computer. Another interesting aspect is that queries can be lifted as well! Allowing to make use of distributed compute and caching (perhaps on pair with incremental compute that is (to be?) built-in into IPVM, i.e., queryResult(oldQueryResult, novelData). And I love how Rhizome graph expressions could make use of other WASM expressions (for photo cropping, etc.). Also I love how provenance info gets lifted to IPVM level / AST level (when an expression has been published (dt), author, who computed). And a side bonus, users can have access to provenance info from within Rhizome queries. This idea of having content-addressable AST is akin to Unison lang approach (which uses content-addressable AST as the source-of-truth, sparing us from having to manage our programs as a bag of text files), but taken up a notch, as we have an expressive graph language perhaps sprinkled with other WASM computations that could have been written on any of those 40+ langs out there, provenance and other cool stuff. It's nice how one could add arbitrary metadata to expressions, e.g., temporal info (giving those neat temporality features offered to us by SQL 2011, XTDB). The devs would be spared the need to track manually "createdAt", "updatedAt", versioning, authorship - it's all in the box! 🔥 On a side node, I love how IPVM manages side-effects for you! Resembles the idea of Algebraic Effects (for reference, found in Unison lang). Making it even more handy for devs. And another topic I wonder about, if I may have your attention for an extra bit, could the ideas of Semantic Web be of use for Rhizome? In particular, there is a graph querying framework, Comunica, that's written in JS and can run in browsers. On top of which there is LDflex, that is a dot.notation interface on top of SPARQL, that aims to make it easy for web developers to interact with graph data. So they can write `db.createUser({name: "Alice"})` to issue a SPARQL transaction, `await db.user` to get data back, `db.user.email.set("[email protected]")` to set an email and so on. Perhaps something like this would be of value to give to Rhizome users to ease the learning curve? There is also a GraphQL interface (GraphQL-LD), for those that want the middle ground between ease of use of dot.notation and expressive power of SPARQL. Another potential value that Semantic Web ecosystem has to offer is RDF1.2, which landed just a couple of months back, and with it there is support for an edge to point to another edge, at last! So edges/tuples can be accreated with any meta we wish, such as "Causality"/which AST expression a tuple came from, temporal info and so on. (and we have access to this meta info from within a query, yay!) Perhaps yet another value is to get the vision of Tim Berners-Lee, and his project Solid, onboard. He's after to make Semantic Web personal, allowing users to have their personal RDF stores (Solid PODs). It's vision aligns with that of Rhizome. Perhaps these two efforts can be aligned? Having Solid on top of a robust data layer, such that of WNFS, would benefit it tremendously, imho. That is what Rhyzome nailed dead right. And having TBL onboard with Fission may give additional weight to Fission's vision. Pardon me for a rather lengthy message 😅, I'm getting all too excited about the current developments of IPVM and Rhizome, and the bright future of the Web they're about to make a reality. I would love to hear your thoughts and carry on a discussion. Perhaps KZbin comments is not the best medium, please let me know of a better place. And thanks again for all your effort into these ground-breaking concepts, and having them developed in the open, for all the people. ♥💪 Rock on, Fission folks! 🤘

Absolutely amazing! I've always wanted something Spreadsheet inspired to run my small "mini programs" inside of