261–269 of 269 posts
Earlier quoted context omitted.
This is the thing with errors and automation. A 1 % error rate in a human process is basically fine. A 1 % error rate in an automated process is hundreds of thousands of errors per day. (See also why automated face recognition in public surveillance cameras might be a bad idea.)
Isn't this just saying "humans are slow" in a different way?
Earlier quoted context omitted.
Human error rates are also not a constant. If you're about to publish a career-ending allegation, you're going to spend some extra time fact-checking it.
Can you point to where that claim was made? I can't find it. The parent post assumes 1% for the sake of argument to underline that the impact of the 1% error depends on the number to which the 1% are applied — automation reduces the effort and increases the number. Hypothetical example: Cops shoot the wrong person in x% of cases. If we equipped all surveillance cameras with guns that also shoot the wrong person in x%…
I'm also making the point that a human with an error rate of x% is not directly comparable to a machine with x% error rate, just via a different line of reasoning.
Pardon my naivetée, but what exactly is JSON-LD doing that the HTML meta tags don't do already? My blog doesn't implement JSON-LD but if you link to my blog on popular social media sites, you still get a fancy link.
JSON-LD / RDFa and such can use the full type hierarchy of schema.org (and other languages) and can build a tree or even a graph of data. Meta elements are limited to property/value pairs.
Either way it sounds awfully expensive for data that probably isn't used by the client most of the time. Do you have to explicitly ask for it? Is there some ad-hoc way to tell the server "hey I don't need the JSON-LD data?"
Earlier quoted context omitted.
I am not sure I understand the fixation on a "killer app" in the context of web standards. We are talking about things like, say, XML, or SVG or HTTP/2. They can have their rationale and their value simply by serving to enable organic growth of a web ecosystem. I think I agree most with your last sentence and should define success more in those terms, aspiring to a better web.
The idea (or hope) is that apps based on semantic standards would kick off a virtuous cycle where publishers of information keep investing in both generating metadata and evolving the standards themselves. As many have mentioned in the thread, thats not a trivial step. People sort of try. A concrete example are the Activitypub/Fediverse standards which dared to use json-ld. To my knowledge so far the social media exp…
Earlier quoted context omitted.
The semantic web has been, in my opinion, a category error. Semantics means meaning and computers/automated systems don't really do meaning very well and certainly don't do intention very well. Mapping the incredible success of The Web onto automated systems hasn't worked because the defining and unique characteristic of The Web is REST and, in particular, the uniform interface of REST. This uniform interface is wast…
> Maybe this all changes when AI takes over, but AI seems to do fine without us defining ontologies, etc. If you say "AI" in 2024, you are probably talking about an LLM. An LLM is a program that pretends to solve semantics by actually entirely avoiding semantics. You feed an LLM a semantically meaningful input, and it will generate a statistically meaningful output that just so happens to look like a semantically mea…
The semantic web standards are sorely lacking (for decades now) a killer application. Not in a theoretical universe of decentralized philosopher-computer-scientists but in the dumbed down, swipe-the-next-30sec-video, adtech oligopolized digital landscape of walled gardens. Providing better search metadata is hardly that killer app. Not in 2024. The lack of adoption has, imho, two components. 1. bad luck: the Web got…
Earlier quoted context omitted.
Gold is and has been a decentralized currency for a very long time. It’s mostly just very inconvenient to transport. > Then we abandoned it in favor of centralized currency for some reason. I don't know, reliability perhaps? The global economy practically requires a centralized currency, because the value of your currency vs other countries becomes extremely important for trading in a global economy (importers want h…
> Gold is and has been a decentralized currency for a very long time. It’s mostly just very inconvenient to transport. Even precious metal coins became endorsed by one authority or another (the cities/banks/little kingdoms stamping the coins). Because you as a normal person don't have the resources to validate every single piece of gold/silver you are paid with. There has also been a short period when every 3rd bank…
It would have been an inconvenient currency for small transactions, but it’s still a currency.
The bank currencies were weird. Iirc, some of that was wrapped up in the Civil War and the Confederate currency being “official” but also basically worthless towards the end of the war. I think the Great Depression killed them, when banks became insolvent and their currencies became worthless.
Earlier quoted context omitted.
> Maybe this all changes when AI takes over, but AI seems to do fine without us defining ontologies, etc. If you say "AI" in 2024, you are probably talking about an LLM. An LLM is a program that pretends to solve semantics by actually entirely avoiding semantics. You feed an LLM a semantically meaningful input, and it will generate a statistically meaningful output that just so happens to look like a semantically mea…
I would happily bite on that; I mostly deal with archives, libraries, museums and how they deal with people and communities. Because of that there is a ton of nuance when it comes to identities (there is a lot of gradation in meaning between "African American" and "Black" or gay and homosexual for example). Things that seem simple are often very complicated and I've spent a good deal of my PhD work working on that (t…
First of all, what's the problem? Computing human-written text.
What's the problem domain? Story. In other words: intentionally written text. By that, I mean text that was written to express some arbitrary meaning. This is smaller than the set of all possible written text, because no one intentionally writes anything that is exclusively nonsensical.
So what's my solution? I call it the Story Empathizer.
---
Every time someone writes text, they encode meaning into it. This even happens on accident: try to write something completely random, and there will always be a reason guiding your result. I call this the original Backstory. This original Backstory contains all of the information that is not written down. It's gone forever, lost to history. What if we could dig it up?
Backstory is a powerful tool. To see why, let's consider one of the most frustratingly powerful features of Story: ambiguity. In order to express a unique idea in Story, you don't need an equivalently unique expression! You can write a Story that literally already means some other specific thing, yet somehow your unique meaning still fits! Doesn't that break some mathematical law of compression? We do this all day every day, so there must be something that makes it possible. That thing is Backstory. We are full of them. In a sense, we are even made of them.
We can never get the original Backstory back, but we can do the next best thing: make a new one. How? By reading Story. When we successfully read a Story, we transform it into a new Backstory. That goes somewhere in the brain. We call it knowledge. We call it memory. We call it worldview. I call this process Empathy.
Empathy is a two way street. We can use it to read, and we can use it to write. When two people communicate, they each create their own contextual Backstory. The goal is to make the two Backstories match.
---
So how do we do it with a computer? This is the tricky part. First, we need some fundamental Backstories to read with, and a program that uses Backstory to read. Then we should be able to put them to work, and recursively build something useful.
I envision a diverse library of Backstories. Once we have that, the hardest part will be choosing which Backstory to use, and why. Backstories provide utility, but they come with assumptions. Enough meta-reading, and we should be able to organize this library well enough. The simple ability to choose what assumptions we are computing with will be incredibly useful.
---
So that's all I've got so far. Every time I try to write a real program, my surroundings take over. Software engineering is fraught with assumptions. It's very difficult to set aside the canonical ways that software is made, and those are precisely what I'm trying to reinvent. I'm getting tripped up by the very problem I intend to solve, and the irony is not lost on me.
Any help or insight would be greatly appreciated. I know this idea is pretty out there, but if it works, it will solve NLP, and factor out all software incompatibility.
You can see what I mean learning about the Solid Protocol, I gave a talk about it a couple of years ago: https://m.youtube.com/watch?v=kPzhykRVDuI