Show HN: An attempt to assess truth quantitatively https://ift.tt/pKunfDJ

Show HN: An attempt to assess truth quantitatively Show HN: An attempt to assess truth quantitatively Everyone has a rough idea of what is true based on their experience and interpretation of available evidence. However, it is hard to assess how true something really is because different people have different experiences and interpretations of the world around them, and because the evidence – or one’s interpretation of that evidence – can change at any moment. Marqt.org attempts to solve this problem by aggregating the wisdom of the crowd to arrive at a quantitative measure of how true something is, and how it might change, in real-time. The project imagines what it would be like if you built an open-source knowledge base like Wikipedia using the format of Twitter and verified everything with Stack Overflow. I initially announced Marqt.org several weeks ago here: https://ift.tt/JHe76fG After hearing from some of you, I have reassessed my assumptions and made some changes that I hope will improve the system and encourage you to try it out. * Privacy and anonymity Against conventional wisdom, I intentionally did not install analytics or tools that track user behavior from the start. I am leaning into this further, allowing you to use the full feature set anonymously. I still do encourage you to sign up for an account though, and there are now further privacy protections for those that do. You can read the details of how I do that in the comment below. * Dialing in how true or false a statement is Truth is not black and white, and while the purpose of Marqt.org is to navigate the nuances around partial truths (by saying something can be 68% true, for example), forcing users to take a binary position to get there doesn’t really seem fair. So now, instead of just being able to “marq it” true or false, you can dial in how true or false a marqt is. This also allows abstaining, by setting your marq to 50% (it will snap to 50 from 45-55). It also somewhat acts as an “undo,” if you marq a marqt by accident. Hotkeys still allow you to go fully true or false with “j” and “k” (or "t/f"), but now you can also hit “n” to go neutral. * It is easier to evaluate than create Adding a remarq puts a lot of pressure on “getting it right” and being comprehensive. It also requires work and time and is something that generative AI can do easily. So now, when you make a marqt, remarqs that argue each side are auto-generated, hopefully providing a contextual base to critically engage with. Then you can upvote or downvote the remarqs based on how remarqable or unremarqable they are, and add a remarq yourself if you want to add to the conversation. * Leaning into the subjectivity of truth I still don’t know what makes a good marqt yet. It took the internet some time to figure out what a good tweet is, so I’m hoping the same can happen with Marqt.org. One of the core assumptions that the marqt is built on is around the subjectivity of truth, and so I’ve front-loaded some marqts that lean more into individual experience ["I am skeptical of most things I see.", "I feel optimistic about the future of humanity.", "I am confident my job cannot be replaced by AI.",] (see below for how marqts are sorted). Lastly, I concede that this project can be seen as a naive idea built on quixotic fantasies, but I genuinely believe that we can solve the problem of misinformation in the age of LLM hallucinations, social echo chambers and media bias, and I sincerely hope that you and Marqt.org can be a part of that eventual solution. arthur@marqt.org https://marqt.org June 13, 2023 at 07:59PM