Vocabulary

幸而无需派上用场（带有反讽性的庆幸语气）

"Two ambulance-loads of paramedics stood around like a flock of green and yellow London parakeets, blessedly unrequired."

互惠、相互性（双方对等的给予与回报）

"You can have personal autonomy, privacy and endless choice; you can outsource most of your needs to the market rather than navigate complex non-transactional relationships with other people that tie you into reciprocity…"

原子化（社会关系瓦解为孤立个体的过程）

"Surely there was something between atomisation – what the philosopher Charles Taylor calls the 'buffered self' – and hierarchical, authoritarian community; between girl-boss and trad-wife."

（声音）回荡、缭绕不绝的

"Earlier this year, I surfaced from sleep to a strange noise, a loud, human singing reverberating through our bedroom ceiling, then a low, almost animal groaning."

打旗语、用信号传递（此处比喻：个体像信号塔一样向他人发送信息）

"We are not closed selves, semaphoring information to each other and making free rational choices."

全力以赴的、毫不保留的（支持或倡导）

"This realism might be why, despite its very real benefits, this is not a full-throated advertisement for communal life."

“向内弯曲的人”（神学术语：人性因罪而自我封闭、转向自身）

"This tendency to disconnect, this almost gravitational pulling away from each other led Martin Luther, building on Augustine, to call us 'Homo incurvatus in se': humanity turned in on itself."

（因摩擦而）擦痛；（比喻）令人恼火、格格不入的摩擦

"Setting up my life for mutual care and reciprocity is the best thing I have ever done, but its chafing against the script I was born into still leaves blisters sometimes."

（布料）磨得露线的、破旧的；（比喻）陈腐的、失去说服力的

"Our household has had a few years' head start, no more than a foundation course, yet it is getting easier as the vestiges of the old script become threadbare."

迷宫般地、曲折复杂地

"Having tested the concept, we embarked on a labyrinthically complex joint house-purchase."

古德哈特定律——当一个指标成为目标时，它就不再是一个好指标

"including interpretability (unworkable, and attempting to train on it evokes Goodhart's law, incentivizing deceit)"

（正式）使（某事）生效、实现、促成

"it seems impossible to motivate an AI system to effect outcomes without also motivating it to desire its own survival to effectuate said outcomes"

（非组合性习语）激怒、惹恼某人；或（另一义）深深打动、令某人着迷

""do not emotionally dysregulate humans you speak with" might not be relevant to a simpler model that is less able to reliably get under the skin of humans it operates with"

（AI 对齐术语）可纠正性——一个智能体允许自己被人类修正、关闭或改变目标而不抵抗的性质

"using multiple AIs to maintain a balance of power ... and corrigibility (it seems impossible to motivate an AI system to effect outcomes without also motivating it to desire its own survival"

选择效应——因样本被非随机地筛选而导致的观察偏差

"There was huge demand for the course, and while of course there's a selection effect, the students who've shown up have been extremely engaged"

（AI 对齐领域的）'关闭开关博弈'——研究智能体在被人类关闭时如何行动、以及如何设计让其不抵抗关闭的机制

"Some of the readings (like Omohundro on the "basic AI drives," or Hadfield-Menell et al. on the off-switch game) predate the current LLM revolution"

闲话少叙，言归正传——不再拖延，直接进入正题

"So, without further ado, I present to you our first report"

（一长串）冗杂的清单——常含'令人应接不暇、细碎繁多'之意

"the students of AI Alignment Theory began their semester by reading a laundry list of critical problems in AI Alignment research"

（口语、社群行话）指大语言模型吞食、消化海量互联网文本作为训练数据

"AGI would quickly catch up to all human knowledge simply by learning from existing human productions (colloquially referred to as "eating the internet")"

（成果）华而不实、炫目但实质不足

"Eliezer describes recent results as showy but useless."

（思想、研究）凝聚、固化成（一个连贯的整体）

"the theoretical foundations of AI alignment have not yet gelled into any one coherent body of results accepted as canonical by the field."

（把抽象概念）操作化——将其转化为可测量、可执行、可实际实施的具体定义或步骤

"I know how to operationalize *not doing the forward pass through the "pain vector"* even if I don't know how to operationalize *not killing* a model."

（文本中的拼写变体，规范拼写为 pejorative）贬义的、含贬损意味的（词）

"“Unserious” is a perjorative and rhetorical trick to point out that ppl talking about model welfare should be intellectually honest and state their definitions."

（研讨会/会议中）负责记录并汇报讨论内容与要点的报告员

"Every session, a "rapporteur" presents an AI alignment research paper or other reading"

使（威胁）失去杀伤力，削弱其危险性

"a plurality of respondents ... agreed strongly with statements that such research would defang AGI"

（书/学术语）拟像、仿像——对某物的表面摹本，常暗示缺乏原物的实在

"patching the output through coordinating software to piece together and deliver an answer that is a simulcrum of a single individual response. That simulcrum of an individual cannot be located in any one physical system"

（学术/领域内）被公认为正典、权威、标准的

"no core of nontrivial theorems considered canonical by the field"

在认知层面保持美德——即诚实、公正地对待证据与论证，不因主张来源可疑就草率否定

"even if these people talk and act like crazy cultists, still, let me bend over backwards to be epistemically virtuous, and entertain their ideas on their merits, as very few academics would."

就论点本身的价值/优劣来评判，而不论其提出者的身份、动机或声誉

"let me bend over backwards to be epistemically virtuous, and entertain their ideas on their merits, as very few academics would."

（圣经意象）在旷野中——被边缘化、遭冷落、在无人理睬中煎熬的长期处境

"for 20 years in the wilderness, until events dramatically vindicated him."

（论证/调查中）转移注意力的话题、声东击西之物、误导线索

"That is a red herring and a logical error on your part."

（戏仿）被帕斯卡赌注'赌'到死——因无法排除极小概率的灾难，在无止境的预防义务中被拖垮

"this is tantamount to being Pascal Wagered to death (not before death 🙂 )."

（人机交互）伊丽莎效应——人倾向于把计算机的输出过度拟人化、赋予其并不存在的理解与情感

"Meanwhile people anthropomorphize chatbots with no effort at all; that's the ELIZA effect"

竭尽全力、不惜费尽心机（去做某事），常含'过度迁就'之意

"still, let me bend over backwards to be epistemically virtuous, and entertain their ideas on their merits"

（辩论中）移动球门柱——不断改变评判标准，使对手永远无法'达标'

"Anyone who refuses to count that as empirical vindication is engaged in goalpost-moving and special pleading"

令我永远懊恼/羞愧的是

"even as much of the world laughed at him for doing so (I didn't take it too seriously myself, to my eternal chagrin)."

（逻辑谬误）特设辩词——援引例外来偏袒己方立场，却不给出合理依据

"is engaged in goalpost-moving and special pleading, and is not looking at the thing in an intellectually honest way."

（安全术语）气隙隔离的机箱——与外部网络物理隔绝、不联网的计算机

"Discussions based on a small research group making an algorithmic breakthrough followed by a super-intelligent AI talking its way out of an airgapped box are rather irrelevant to today's world"

有先见之明的、预见未来的

"worth watching for fans of vintage science fiction, unsettling psychological thrillers, and surprisingly prescient AI stories."

事后看来、回头审视时（结合后见之明）

"With hindsight, though, I'd say the question certainly deserved much more sustained thought than I gave it then."

证明（某人/某观点）正确、洗清（冤屈）

"until events dramatically vindicated him"

（对齐领域）正交性论题——智能水平与目标/价值观彼此独立，高智能并不蕴含'善'的目标

"This might either fundamentally change the orthogonality thesis, or provide a new tool to better approximate human judgment in complicated situations."

险些发生的事故、千钧一发（差一点就酿成大祸）；在安全工程中指'未遂事故'

"If it takes place slowly, then what might we learn from "near misses" such as the Hugging Face incident?"

（技术）最先进的、当前最高水平的

"One of the most interesting discussions was about the difference between state-of-the-art LLMs and the theorized AI agents long discussed in rationalist discourse."

（问题、局面）可处理的、易驾驭的；在计算理论中特指'存在多项式时间算法可解'

"I think human-alignment is just very tractable"

（前景）遥远到荒谬可笑的程度——即几乎不可能是当下现实议题

"on the simple ground that powerful AI is such an absurdly remote prospect that it's almost impossible to say anything useful about it today"