Context so far - Work to open up the forest of what has yet to be written. - Semantic Annotation as âTax Payment - Cryptocurrency as payment for work done for AI
nishio It just occurred to me that if a structure is created in which a huge LLM pays people who give it âuseful information that we donât know yetâ, it is too utopian to assume that people will be engaged in creating new information. I thought that âpeople will be engaged in creating new informationâ is too utopian an assumption, and that there will be a large number of âpeople who will drop secret information to LLMs and turn it into moneyâ before that happens.
-
nishio Is it utopia because we can evolve into an integrated thought body where secrets and lies do not exist? ()
0xtkgshn How do you determine what you donât know yet? Iâm wondering if you monitor the words that people put in, respond anyway, and let them rate the responses, and the low ones are the âsought-afterâ ones, and you give back a reward for putting in the domain knowledge there.
-
nishio Itâs relatively easy to determine âwhat we donât know yetâ and âwhat is requiredâ but difficult to determine âthat the answer is not falseâ. Basically, I think you have to collect a lot of them and throw out the outliers.
-
nishio wanted: often asked by humans, but not well received for answers
-
What we donât know yet (what we didnât know yet): that we canât output the given information by generating a question text that makes the given information the answer and letting the old version answer the question.
-
-
0xtkgshn peer prediction method feeling (the very essence of crowdsourcing)
-
0xtkgshn Slightly off topic, but I just remembered that the prediction market is not just a transaction that benefits the end user, but by looking at the convergence point of that transaction I remember that you can gather information by looking at the convergence point of the transactions. The information you get from the market is more accurate than that of a normal, well-meaning user.
-
-
nishio Ah, but the style of discarding the outlier would play âa truth different from most peopleâs expectationsââŠ
-
-
nishio Well, sure. If the information is false, it would be ability to deter (an attack, etc.) against those who input appropriate information, and if a human decides the amount of wager, can it be regarded as a statement of confidence?
-
0xtkgshn Staking is really profound, Vitalikâs knowledge of game theory and social science is remarkable! Vitalik]âs knowledge of game theory and social science is remarkable.
-
-
(Own note)[/tkgshn/The P + epsilon Attack](https://scrapbox.io/tkgshn/The P + epsilon Attack).
-
sirouto Iâm sorry to ask a simple question, but would you pay a fee when you can learn for free from the internet?ăEven if you buy unknown information, you need credentials. For example, information that âCompany Aâ will launch a new product next year cannot be determined to be true by machine learning of text. Rather, it is likely to take a posting fee, similar to search, and place a link ad in the AIâs response.
-
nishio This is after the information that can be learned for free from the internet has been learned and it becomes a race to find out how to get other information. There is no need for AI to do the true/false judgment, just give it to humans and let humans make the judgment on whether it is useful or not.
shima__shima Before that happens, I think there will be a lot of people spouting off random things.
-
nishio If it is determined that you told a lie against other information, your account will be banned and you will die financially ability to deter (an attack, etc.) will occur.
This page is auto-translated from /nishio/ç§ćŻæ ć ±ăLLMă«ćă蟌ăă§éă«ć€ăăäșșăźć€§éçșç using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. Iâm very happy to spread my thought to non-Japanese readers.