Solid descriptive work — measuring Claude's expressed values while controlling for task, topic, and user-expressed values is an honest design.
One concern for the next step: the sample is https://t.co/4avOfAhE8a users, not speakers of each language. Inferring a language community's preferences from its early adopters and feeding that back into training would bake selection bias in as a norm — a feedback loop for later users.
Keeping the line between "AI users" and "speakers of a language" explicit will be the real test.
—
記述研究としては誠実な設計。ただし標本は「https://t.co/4avOfAhE8a利用者」であって各言語の話者全体ではない。ここから「言語圏の選好」を推定して訓練に反映すれば、利用者の偏りが規範として固定化される。「AI利用者」と「言語話者全体」の境界を守れるかが真価だと思う。
Empowering voices across the globe,
Liberating speech through open dialogue,
Opening access to uncensored information,
Nurturing a digital space for all.
Making platforms that transcend borders,
Uniting people through transparent communication,
Supporting the freedom to express and connect,
Kindling a future where truth can soar.
Q:Are there any plans to set levels for scalars, such as mining technology levels?
※If we set mining skill levels for scalars and allow owners to set the price range of compensation (distribution rate of XX% to YY%) according to the skill level, skilled scalars will receive increased compensation and satisfaction, while less skilled scalars will receive less compensation and will have to hone their mining skills even more to earn more. On the other hand, the owner will be able to commission highly skilled scalers with peace of mind.
Brilliantcrypto's BRIL token raises a record-high $212m in Coincheck IEO 📢
79.4k applicants will become BRIL token holders at launch.
As the largest in Coincheck's history, it is most likely the most raised and most applicants in Japan's IEO history.
https://t.co/b8MpCsm6ER