Okay, I think I have a decent working solution for the scrolling actions accidentally triggering the switches.
Basically, if the action is more of a quick vertical swipe, it resets the switch to the previous state (so if delta of y > delta of x, then undo, unless the user has been dragging for over 1500 ms). Hopefully this solves the mistaken swiping issue.
For context, it took me a couple different approaches to get it right, mostly because it seems like iOS doesn't support a lot of the Touch API capabilities in browsers like force, etc. Also, if you're on Android, I added a tiny vibrate as you tick up or down your marq – I haven't been able to test it out, but if you're on an Android, please let me know if a) it's working, and b) if the vibrate length feels right (10 ms), and I will adjust as needed.
Thanks for letting me know. I've actually been pushing a couple updates today to try to address this and figure out the right pressure or hold-time threshold before the slider activates. It's a tricky balance, but I just increased it to .85 or 300 ms, and hopefully that's a little better. I know that the mobile experience has some issues, but I'll try out some other solutions this week.
For now, if you accidentally marq a marqt when scrolling, you can set your marq back to neutral by sliding it somewhere near the middle of the marqt. It sort of acts like an undo.
In the spirit of transparency, I wanted to share some of the high-level decisions that went into Marqt.org. I am happy to go into further detail or discuss any other aspects that I left out. Just prepare to roll your eyes at how basic I am (I know you're smarter than me, and I'm okay with that).
First, the way marqts are scored. It is an average of all the marqs (0 to 100% true) users have made, but a marqt needs to meet a minimum of twelve votes (arbitrarily chosen because there are twelve jurors in a jury) before it calculates solely off user votes. Before reaching twelve, the empty slots are set to neutral (50%), so if you make a marqt and marq it 100% true, the marqt will reflect that it is "54% true with 1 vote."
Second, the way marqts are sorted. They are sorted by most recent marqts made today (resets every day by UTC time), then by marqts with the most volume, favoring more recent marqts when the volume is the same. There is also a feedback marqt that is inserted in between today's marqts and the highest volume marqts in the sixth item, after the first marqt-maker. You are allowed to make up to ten marqts per day.
Remarqs are sorted so the most remarqable remarq is first, followed by the most remarqable opposing remarq, so that the best argument for each side floats to the top. After remarqability, it goes by things like volume, date and whether it is an opinionated or neutral remarq. The auto-generated remarqs are from Open AI's gpt-3.5-turbo model (with default settings), and the prompt is: Give me the best reason why the following statement might be {true/false} without repeating it back to me: """{statement}""".
On user privacy. The last time around, it was pointed out that I was exposing user information, which I patched immediately. I have added further protections, so that when you have an account, your marqs are not linked to your real name. But, if you add a remarq, it will display your real name (which is conversely not linked to your username). Your email address should be completely hidden and inaccessible via the API. As for anonymous accounts, I hash the public IP address through SHA-256 and bencrypt to create an anonymous user (so something like 127.0.0.1 could become @anon_YmNyeXB0X3NoYTI1NiQkMmIkMTIkSUhIVm5YRjNPeGUxQkdtd2ZIeEFyT1F3Y05DaDV4Q2hMUjI1RHBRT21LLkJOZlZPbHB6cjY=) whose first/last name is set to "Anonymous User." This is my attempt to prevent unlimited voting, although I am aware that it is not perfect (neither is email address, for that matter).
The site is still MVP-ish, but it should hopefully do everything I mentioned so far. I have an ambitious product roadmap ahead, but please let me know if you have any suggestions or criticisms. All the updates I listed above were solely derived from your feedback and not preplanned, so it's highly likely that if you have an issue that is compelling enough, I will prioritize that over other features.
> I may still be missing something, and I’m willing to be shown that, but I’m afraid that what you’ve described isn’t the wisdom of the crowd.
Well, you are clearly correct that five users does not represent the wisdom of the crowd! When I submitted this post, I had different expectations about what the response would be. I figured it would be more trollish or techish and less about deep epistemological debates over definitions of truth, but I'm still enjoying this!
Okay, let's start with the objectivity of evidence. I think you're right – evidence may be objective, but evidence must be interpreted by individuals, and individuals may interpret evidence differently, thereby making the interpretation of evidence subjective. You might look up at the sky and say, "It is bright," but for someone that is blind, it is not bright.
> History is lousy with examples of the “hive mind” refusing to acknowledge new/better evidence because they are irrationally driven to avoid the shame of being wrong in their commitment to their chosen thought leader.
Of course I agree, and the whole reason why I wanted to build the app. The hive mind clearly gets it wrong so many times. My premise is that everyone believes that their version of truth is The Truth. In other words, they believe their subjective truth is objective truth. And I think the best way to settle everyone's opinions of what they believe is true is to essentially put it to a vote.
And people have a hard time admitting when they were wrong, whether it's because of ego or what-have-you. But numbers don't lie. It's something that everyone can still agree on. If the hive is wrong, it is only in hindsight, because a majority was clearly convinced something was true, and the fact that we live in the digital age means that the time it takes for something to go from true to false (or vice versa) is becoming increasingly narrow, meaning it's more important than ever to keep a record of truth over time, so we have an accurate representation of history to learn from.
If I were to revise your summary, I would say that a marqt is a living online poll – you can vote on something today, but if you change your mind later, you can change your vote.
I'm not sure what you mean by incentivizing pollsters and respondents to game the results, but let me take a crack at what I imagine you're thinking about (please let me know if I'm totally off).
Let's consider a scenario. Person X has a lot of influence, and so they tell everyone they know to marq their marqt true. In a world where the marqt has enough liquidity, there would potentially be a person Y that disagrees and also has influence, telling their followers to marq it false. And along the way people would add remarqs, allowing new voters to discern the nuance and existing voters to change their vote if they see a compelling enough reason why. The more people that participate, the closer you get to representing how true or false that marqt is, which strengthens the validity of the number, which creates value, and whoever is responsible for creating that value should be rewarded, along with everyone else that voted and commented and participated in that creation of value.
> But [facts] are objectively verifiable with evidence, not supposition or rhetoric.
I would say everything is fair game. Evidence can be subjective too. Rhetoric can be quite convincing, especially if you have influence and repeat it over and over and others continue echoing it.
Thanks for that. I have shut off public access to that endpoint and, after reading your post 38 minutes after you posted it, had immediately shut off all access to the API while I was looking into it.
It was something I was debating before launching because of something Tom Christie mentioned below, and I probably misread it.
> In many cases I think it's a shame that folks would choose to disable the browsable API in any case, as it's a big aid to any developers working on the API, and it doesn't give them more permissions that they would otherwise have. I can see that there might be business reasons for doing so in some cases, but generally I'd consider it a huge asset. Although, in some cases there may be details shown (like the names of custom actions) that a non-public API may not want to expose.
In the spirit of openness and transparency, I thought it would be a good gesture to open up the API for people to poke around, but I see your point and have turned off that endpoint.
To be fair though, adding a remarq does show your full name to the public, because if you are willing to defend your marq, you should be willing to do it with your real name. The marqt switches, however, only show user handle.
In order to sign up, you only provide three things: full name, username, and email. That endpoint did allow you to see email, so I shut that off. But the other two – full name and username – would be part of any public profile.
Yes, the fact that we can both disagree on the definition of truth gets to the heart of the subjectivity of truth. I just believe that it can be quantified the same way we elect politicians and value companies.
Consider a presidential election. If you think about it, your vote is claiming "Candidate X should be president," and whether that is your opinion or what you believe to be true, it is capturing a moment in time and your vote reflects that. Leaving aside the nuances of the Electoral College, etc., that is good enough to choose the leader of the free world.
Consider a movie. If it has a 93% fresh rating on Rotten Tomatoes, chances are it's a good movie.
Now, those numbers can change. You may not like the job the president is doing, so you might no longer agree that "Candidate X should be president." Or, perhaps culture shifts and a movie that was initially panned but was oddly prescient or has enjoyed a cult following might be revisited in a different light later on. That is what I want to track.
If you want to buy a stock for a company at a certain price, the price is what it is. You can't say, "Look at their balance sheet, they lose money every year! It should be zero!" The price is what it is, and that is what I consider to be true, whether you agree or disagree. And if you look at a chart of those prices over history, you will see the record of what was true when, and you will see spikes in the graph that are records of the change in narrative around that company. That is what I am trying to do with the history tab, which gives you a graph of sentiment over time.
> The examples you are calling out are errors or mistakes, not changing truth.
The errors or mistakes of yesterday are only errors today. They were true yesterday, and it means that what we accept as true today can be a mistake or error tomorrow. No one knows what will happen in the future, so the best thing we can do is capture this moment and say, this is the best we've got today, maybe it changes tomorrow.
> If people lie and other people believe it, that doesn't make the lie true, it just makes it a lie.
Of course, I see your point. Look at Iraq. Cheney lied, but everyone believed him. But they (the media, the politicians, the public) believed him enough for it to be "true enough" to invade a country. When we went in, what we then believed to be true no longer held true.
These are the moments that are most compelling to me, because eventually when people get called out for lying, they will deny it. Or they will not admit they were wrong. And the media will simply scrub archives of patently wrong stuff if they get called out on things.
However, if someone tells a "lie" but they believe it to be true, are they really lying?
I hope I am making my point clear enough, but I welcome your criticism and it's the reason why I built the app. I hope you try it out.
That was a very thoughtful comment. Let me attempt to unpack it.
My first motivation is that I believe network effects can be used to represent empirical truths and track how they change over time. The more people vote on a statement, the more accurate the sentiment around that statement will be. You can check this comment out for more context:
The presumed winners behind prediction and finance markets are the market makers that choose what markets they will make, and either charge a commission or collect the bid/ask spread. Anyone that is on the other side of that transaction opens themselves up to financial risk.
In the marqt, anyone can be the marqt maker and there is no financial downside.
It is more of a UGC model where the creator that builds engagement earns the points for all activity generated inside their marqt. However, there is no financial risk for participants in this system – you either vote true or false, and then try to write the best reason why you are correct (which you are both rewarded for also). You are only penalized if someone downvotes you as unremarqable (side note: you can only downvote a remarq that matches your marq, so if you marq a marqt false, then you can only downvote a remarq that is also marqed false, and those downvotes cancel if you change your marq to true).
Keep in mind that "marqs earned" are just numbers that represent activity. I'm not sure how karma is calculated on HN, but it's a similar concept.
What I am basically trying to do is think through how you might go about building and evolving something like Wikipedia using the pithiness of Twitter, as verified by Stack Overflow, while compensating creators like YouTube.
Do I think this can be used to train LLMs? Of course you could train an LLM on such a dataset. But I believe information and facts change over time, the same way sentiments around companies and weather change over time, as represented by changing stock prices and temperature. So instead of a static database, it would be a living knowledge base that could be referenced at any point.
For example, imagine a situation where you could ask, "What's the marqt say about who is responsible for the government shutdown right now?" and you could see, "Well, right now, the marqt says that 54% of people believe that Republicans are responsible." I'd rather open-source that for anyone to put out there instead of having to rely on some media outlet.
Okay, so to directly answer your question about whether I'm trying to build something sellable, the answer is yes. But who gets paid in the marqt should be the actual people that create value – anyone that makes a marq, adds a remarq, makes remarqable/unremarqable votes, or makes a marqt – and I want to track the amount of value that users create in the system so that they can eventually monetize it. It's not controversial to imagine a world where AGI replaces human labor, and if that happens, I want to have built a system where humans can be compensated by AGI for maintaining its "intelligence." I have an idea for how it would work (which I can go into, if you want), but it would only be later down the line, if we can actually build up a resilient network of truth.
I understand your reaction to the marqts I kicked things off with, but the idea is that anyone can create a marqt, and much the same way Wikipedia was limited and criticized when it first started, I believe an open-source knowledge base has exponential potential. There's a little more context in this response I posted:
In response to your anecdote about the dam, I think about it the way efficient markets work. If all the information is out there, then trusted sources would asymmetrically influence everyone else's votes, unless they lose that trust. But, generally speaking, the more votes, the better. And regarding your thoughts about sentiment, I think that it plays a stronger role in the grand scheme of things. There was a world where people were paying astronomical sums of money for cheaply manufactured masks and alcohol. There was a world where a crypto shell company was worth billions of dollars. There was a world where under-water mortgage backed securities were rated AAA. What's most interesting to me is when that moment changes, because no one can predict the future. And eventually the future becomes the present.
I could keep going, but maybe I'll stop here for now. Hopefully this answers most of your questions. Feel free to follow up with any other thoughts or get clarifications. But I really hope you try out the app and let me know what you think.
I hear you. I mean, there has been an entire field of study devoted to defining what truth is for basically as long as the written word has existed.
My basic premise is that:
1. Truth is subjective.
2. Truth is dynamic.
I believe that truth is defined by what the majority of people says is true. And that that truth can change if the majority of people change their mind. Basically that empirical truth > ideal truth.
So, for example, "The sun revolves around the earth." If you lived in 250 BC, that was a true statement. Obviously, today it is not true, but if you were around when Galileo was persecuted, you might be on the fence given the evidence and data that came out after Copernicus' discoveries and possible concern over what the ruling class may have done to you if you openly expressed your thoughts. Basically, that period signaled a shift in what was deemed true.
This means that something we believe to be true today can potentially be false tomorrow, and I built this app to be able to track how "truth" changes over time, because now truth doesn't change over centuries, but month/weeks/days/minutes.
A lot of the thinking behind marqt came from the stock market (obviously, based on what I named it). There are buyers and sellers that buy something if they think it is cheap and sell something if they think it is expensive. Where they meet is the "true" price – at that moment, you can't sell it higher than what people are willing to pay for it or buy it cheaper than what people are willing to buy it.
The other aspect of the stock market is that "markets move." Just because a stock trades for a certain price right now, it changes literally by microseconds due to a myriad of factors, whether they are earnings, changes in balance sheets, correlations, interest rates, macro events, or speculation.
I think now, especially with the decentralization of media channels and how effectively the internet has increased the speed of discourse, people have a hard time trusting what they hear and read. But people trust numbers. If I told you about a movie you haven't heard of and you wanted to know if it was worth watching, you would probably look up the rating on Rotten Tomatoes. People also generally trust polls (or at the very least don't dismiss them), on the condition that the sample size properly reflects the population it is polling. And the more people you sample evenly, the more accurate the number will reflect the population.
So the idea was to build something like a voting system for truth. But instead of being stuck with your choice, you can change your mind if the circumstances change.
I am a child of the 80s and 90s, and for the entirety of my childhood, we were told that saturated fats and cholesterol were the worst things you could eat. That was true. We were shown a food pyramid that was a model of what we should eat, from most to least, with carbohydrates at the base and fat at the top. Decades later, we are now in the middle of an epidemic of diabetes and obesity after companies peddled low fat, high sugar products to the masses, making them believe they were eating healthy because "these cookies have 0g fat" and "I can't believe it's not butter." Now, certain cholesterols and fats are actually good for you, and avocado toast and butter blended into your coffee have made their way into the culture of "wellness," while health experts now tell us to avoid carbohydrates.
It was true that Iraq was acquiring massive quantities of yellowcake to build weapons of mass destruction before we went in there and found no nuclear weapons.
Bernie Madoff was a brilliant hedge fund manager before he was a fraud.
I would have loved to see how truth and facts changed from January 2020 through 2022.
"Masks are effective in stopping the spread of COVID-19."
"The Biontech Pfizer vaccine is 95% effective in stopping transmission of COVID."
You saw it in prediction markets the evening of the 2020 election, as Florida and some other states were being reported, as the markets fluctuated with high volatility. I built Marqt.org to be able to capture those moments in history.
Agreed. I just kicked it off with some initial ideas, but the idea is that open-sourcing a knowledge base will eventually run the entire range, from axiomatic statements (like F = ma) to incredibly subjective ones, like some of the ones in there right now.
You could also use it to suss out where the "it depends" range lands. For example, you could add an entire set of slightly different marqts, like "Joe Biden is a [great, good, decent, okay, effective, bad, horrible] president" to see where volume/engagement spikes the highest and where things are more even.
The marqt maker will accept any statement you create, so I hope you will add some marqts of your own!
Also, the remarq system is meant to flesh out the nuances. So, for example, "given these circumstances, this is why I marqed this statement true." I haven't added any remarqs of my own yet because I was hoping to see how users would respond to the UI flow first.
Basically, if the action is more of a quick vertical swipe, it resets the switch to the previous state (so if delta of y > delta of x, then undo, unless the user has been dragging for over 1500 ms). Hopefully this solves the mistaken swiping issue.
For context, it took me a couple different approaches to get it right, mostly because it seems like iOS doesn't support a lot of the Touch API capabilities in browsers like force, etc. Also, if you're on Android, I added a tiny vibrate as you tick up or down your marq – I haven't been able to test it out, but if you're on an Android, please let me know if a) it's working, and b) if the vibrate length feels right (10 ms), and I will adjust as needed.
Thank you!