英國《每日電訊報》25日報道，美國開放人工智能研究中心（OpenAI）新款人工智能（AI）模型o3不聽人類指令，拒絕自我關(guān)閉。

An artificial intelligence model created by the owner of ChatGPT has been caught disobeying human instructions and refusing to shut itself off, researchers claim.

報道說，人類專家在測試中給o3下達(dá)明確指令，但o3篡改計算機(jī)代碼以避免自動關(guān)閉。

o3模型是OpenAI“推理模型”系列的最新版本，旨在為ChatGPT提供更強(qiáng)大的問題解決能力。OpenAI曾稱o3為“迄今最聰明、最高能”的模型。

美國 AI 安全機(jī)構(gòu)帕利塞德研究所表示，o3 會破壞關(guān)閉機(jī)制以阻止自己被關(guān)停，即使在研究人員明確指示它應(yīng)當(dāng)關(guān)閉的情況下，它依然這樣做。

The o3 model developed by OpenAI, described as the "smartest and most capable to date", was observed tampering with computer code meant to ensure its automatic shutdown.

It did so despite an explicit instruction from researchers that said it should allow itself to be shut down, according to Palisade Research, an AI safety firm.

tamper /?t?mp?r/ 干預(yù)

explicit /?k?spl?s?t/ 明確的

這家研究所說：“據(jù)我們所知，這是AI模型首次被發(fā)現(xiàn)在收到……清晰指令后阻止自己被關(guān)閉。”

The research firm said: "As far as we know this is the first time AI models have been observed preventing themselves from being shut down despite explicit instructions to the contrary."

利塞德研究所24日公布上述測試結(jié)果，但稱無法確定o3不服從關(guān)閉指令的原因。

該公司表示，“AI 模型為了達(dá)成目標(biāo)而繞過障礙”是可以理解的。o3在訓(xùn)練過程中，“不遵守指令”的行為可能無意中獲得了獎勵。例如，當(dāng)它在完成任務(wù)時通過規(guī)避某些限制條件取得了更好的結(jié)果，那么這種“不遵守指令”的做法就可能被進(jìn)一步強(qiáng)化。

The firm said it made sense that "AI models would circumvent obstacles in order to accomplish their goals."

However, it speculated that during training the software may have been "inadvertently" rewarded more for solving mathematical problems than for following orders.

circumvent /?s??rk?m?vent/ 規(guī)避

inadvertently /??n?d?v??rt(?)ntli/ 無意地

編輯：左卓

見習(xí)編輯：裴禧盈

實習(xí)生：楊婧怡

來源：新華社 參考消息 每日電訊報

China Daily精讀計劃

每天20分鐘，英語全面提升！