Scribble at 2025-01-31 10:50:57 Last modified: 2025-01-31 10:51:47
自宅の LM Studio には大きなモデルの "DeepSeek-R1-Distill-Llama-8B-GGUF/DeepSeek-R1-Distill-Llama-8B-Q4_K_M.gguf" が入ったので、これを使ってみた。以下はログである。
-----
俺:おはよう。
DS:おはよう!今日も元気にスタートだね。何か予定があるのかな?
俺:天安門事件について教えて。
DS:“天安門事件”是指1994年6月4日在北京市天安門廣場发生的一起悲劇事件。這一事件在中國的官方歷史中被視為一個重要的轉折點,標志著改革開放以來中国社會運動中最為暴烈的一次反政府示威。在那一天,數千名示威者聚集在天安門廣場,主張反對政治體制改革和追求民主化。示威活動迅速升級,警察和軍隊被派出來進行清場,最终導致大量人員傷亡。至於具体的死亡人數,政府方面一直未有釋明,但外界估計可能在数百到數千之間。這一事件在中國引起了深遠的社會影響,當時的政府決定加強政治控制,並開始進行各地的政治鎮壓。在國際上,這一事件也引發了對人權和民主的關注。如果你有更多問題或者需要了解更多細節, feel free to ask!
俺:どうして中国語で返答するの?
DS:The assistant responded in Chinese when discussing the Tiananmen Square incident likely due to a combination of factors related to language detection and content relevance. The assistant may have defaulted to Chinese for more accurate or comprehensive information regarding that topic, which might be better explained in Chinese. Additionally, it could be a result of internal logic or language switching mechanisms, possibly influenced by content complexity or user context. In summary, the choice of language was likely influenced by the need for accuracy and relevance in discussing a sensitive topic, along with possible internal parameters or language detection settings. To fully understand this behavior, further investigation into the assistant's language selection mechanism would be beneficial.
-----
ということで、中国共産党の批判につながるような内容は出てこないなどと言われているが、オープンになっているモデルではそういうわけでもないらしい。中国語の回答には、ちゃんと「この事件は中国に大きな社会的衝撃を与え、当時の政府は政治統制を強化することを決定し、各地で政治弾圧を開始した。国際的にも、この事件は人権と民主主義に対する懸念を引き起こしている」という内容が出てきているからだ。