Search Ideas
4643 ideas match your query.:
I think that you've made something really great here with Veritula, I'm naturally loving it. It would be a shame to make the experience taxing as one must be conscious of AI scores. I looked at the results of the analysis you did on my criticism. It was practically a sentence and a bit, that was adjusted by AI, as I mentioned it was the result of a speech to text tool, I don't type much anymore, it's a part of my workflow (a % score also does not consider the fact that replies are usually short. So one sentence can be a significant)
I think, especially on a platform like this, users don't enjoy feeling restrained. The problem with AI is that, used recklessly, it creates low quality ideas/criticism (many times because it does not actually address a problem so progress stalls). When one does not have a reliable system with which to consider a response as AI, the label and its enforcement becomes worthless. This reminds me of your criticism of hard to vary, in that because we cannot program it we do not understand it. Your rule on AI and labelling seems to suffer from the same issue.
Kitty Comes Home: my first game made with Claude Opus 5.5.
You play as a little kitten trying to find your way back to your mom.
Play for free here: https://blog.dennishackethal.com/posts/kitty-comes-home
Teaser trailer:
As a rule of thumb, please try making your writing less verbose. For example: ‘unknowability implies existence.’
Brevity helps not just your readers but yourself: the longer your text, the more vulnerable it becomes to criticism.
If your readers need more info, they can ask.
I haven’t used Wispr Flow, but their site says it “turns speech into clear, polished writing.” Sounds like it’s not just plain speech-to-text, and the forum rules count AI polishing. Even if I misunderstand Wispr Flow, the forum rules make it your responsibility that AI detectors come back negative.
I’ve retroactively marked #6015 as AI-assisted. It won’t count toward badges. Any replies of yours below #6016 won’t either, since they only exist because of #6015. If it happens again, I’ll lock your account for a week.
I used Wispr Flow, speech-to-text. Tough line to draw.
Pangram estimates about 45% of this idea to be AI-generated. True?
Thanks. I agree that knowing an algorithm and understanding why it is impossible are different things. The theorem says there is no universal halting algorithm to know.
The contradiction arises when we assume that an algorithm satisfying those requirements exists. The conclusion is that no algorithm meets those requirements. Describing such an algorithm as “unknowable” seems to treat it as something that exists but is inaccessible to us, when its existence is precisely what the theorem rules out.
Consider an exact circle made from finitely many straight line segments. We will never discover what such a figure looks like because no figure satisfies those requirements. Calling its appearance “unknowable” seems to turn the nonexistence of an object into a limitation on our knowledge of it.
In the same way, “we will never know a universal halting algorithm” amounts to saying that we will never discover an algorithm that does not exist, based on our current theories. If that is all you mean by “unknowable,” I think the only thing we would not agree about is the definition of “unknowable” itself.
Most links won’t have incoming links themselves.
One thing I haven’t decided yet: each circle in the graphic links to its idea, but it could instead link to its own page of incoming links.
Done as of 14003625. Each idea now has a separate page listing all the ideas that link to it. For example: https://veritula.com/ideas/5546-how-does-veritula-work-veritula-latin-for/links
The page features a ring with discussions and profiles grouping the referring ideas:
The page as a whole looks like this, listing the ideas in text form as well:
Last but not least, an idea with referring links gets a new pill badge:
Let me know if this is along the lines of what you had in mind.
Hi Michael, welcome to V.
The halting problem shows that no algorithm can correctly decide, for every program and input, whether the program will halt. Yet we can understand why such an algorithm is impossible, and we can determine the answer in particular cases.
Yes, we can, but note the switch from knowing such an impossible algorithm to understanding why such an algorithm is impossible. Those are two different things.
What exactly is being claimed to be unknowable here?
A universal halting algorithm.
It is unclear to me why uncomputable should mean unknowable. The halting problem shows that no algorithm can correctly decide, for every program and input, whether the program will halt. Yet we can understand why such an algorithm is impossible, and we can determine the answer in particular cases. Are you equating the impossibility of a universal computational procedure with the impossibility of knowledge? What exactly is being claimed to be unknowable here?
How I find myself using Veritula at the moment is looking at the latest developing threads. Catching up on the context of that thread through clicking “Activity” is not time-effective; it reminds me of trying to catch up with a group chat full of messages. Threads naturally become messy, and not all points serve to advance the problem (some can be misunderstandings or micro-corrections). For this reason, there should be some summary that considers the following as context:
- The thread problem + metadata
- The discussion points under that thread (considering different branches) that relate to the problem.
- A prompt that aims at addressing “messiness” minimisation.
The summary should preserve substantive developments, competing answers and unresolved criticisms, while compressing resolved misunderstandings and linking back to the relevant posts.
If such a setup, including the summary, prompt and rules for what gets included, were kept open and criticiazable, we should arrive at a good system (so there should be a thread where we can iterate on ideas).
"Veritula could be a place where people break news stories and others can quickly fact-check and improve upon reports by revising them."
I ask myself where I go when I want 'answers' to such problems and immediately I think prediction markets. It's mainly because I think the financial incentives and outcome based 'test' serve as highly reliable error correction. I wonder what different utility Veritula would add in this regard.
… if the discussion itself is clear and concise …
Sure but often they’re not…
I might do that too. It arguably already falls under the rule against sabotaging debate. But the rules already explain the need for disclosure as well.
I don’t understand why you’re fighting this rule so hard. It’s not like you’re not allowed to use AI. You are. Just disclose it.
Then prohibit unnecessary verbosity directly. Human-written text can be verbose, and AI-edited text can be concise. That doesn’t explain why someone using AI to edit their own argument should have to disclose it.
Fair, but if the discussion itself is clear and concise, it doesn’t need an AI summary either. I agree that a long discussion could benefit even when each individual idea is concise.
I think you defeated my criticism. That’s kind of cool, I get it.
This could extend beyond fixing errors to improving page load times, search quality, and reducing churn, provided you can measure whether a change helped. My friend Samay co-wrote this about a similar loop using production feedback and evals to improve tax agents:
https://openai.com/index/building-self-improving-tax-agents-with-codex/
Kitty Comes Home: my first game made with Claude Opus 5.5.
You play as a little kitten trying to find your way back to your mom.
Play for free here: https://blog.dennishackethal.com/posts/kitty-comes-home
Teaser trailer: https://youtu.be/bcM2BI-41cY