求职申请要求我提供 SAT 分数
Job application asked for my SAT scores

原始链接: https://mrmarket.lol/job-application-asked-for-my-sat-scores/

作者最近在求职时遇到了一项要求提交 SAT 分数的要求,这引发了他对招聘实践的深层思考。虽然 SAT 分数可能与认知能力相关,但它是一个有缺陷的预测指标,过度依赖申请者青少年时期的表现,却无法考量其随后数十年的个人成长或环境因素。 从历史上看,招聘方式在标准化认知测试、评估中心以及各种古怪且无效的方法(从托马斯·爱迪生的“盐测试”到现代脑筋急转弯)之间摇摆不定。研究表明,结构化面试和工作样本测试仍然是预测职业成功最有效的指标。 然而,生成式人工智能的兴起使得简历和居家作业等传统的远程评估方式日益不可靠。作者认为,企业要求提供 SAT 分数,是在人工智能能够轻易生成高质量、欺骗性申请材料的时代,试图衡量“原始智力”的一种孤注一掷的做法。为了应对这一挑战,作者建议回归到经过验证、受控程度更高的评估方式,如面对面的白板演示、现场角色扮演或监考任务,以确保候选人具备岗位所需的真实技能和批判性思维,而非仅仅是驾驭 AI 模型的能力。

最近的一场 Hacker News 讨论凸显了一个日益增长的趋势:越来越多的公司开始要求求职者提供 SAT 分数,这引发了关于招聘实践和精英管理论的激烈争论。 支持者认为,随着大学学位普遍存在分数通胀现象,标准化考试可以作为衡量“基础流利度”的客观基准。支持者认为,由于智商随时间推移保持稳定,SAT 分数可作为衡量高压工作环境中智力与竞争力的可靠且低成本的指标。 然而,批评者则将这种做法视为一种排他性的“年龄过滤器”或“移民过滤器”,并指出许多国际候选人或资深专业人士从未参加过该考试。其他人则认为,仅凭一个人的青少年时期表现来评判候选人,忽略了其个人的成长与生活境遇。许多评论者还担心,此类指标更看重应试能力和社会适应性,而非真正的创造力,这可能会将那些“古怪”或非传统的天才拒之门外。 总的来说,该社区对此分歧严重:一些人将其视为日益竞争激烈的环境中一种务实的筛选手段,而另一些人则谴责它是一种不可靠且依赖自述的指标,不仅容易助长不诚实行为,还偏袒狭隘、从众的群体。
相关文章

原文

Yesterday I was looking at YC jobs to see if there were any interesting small startups hiring in my area (while I otherwise love my current fully-remote job, I'm looking for hybrid). I found one that advertised an opening they called "GTM". I thought that was cool re: no specific scope, just someone who could make their product attractive to buyers however they can.

Product seemed good (ETL type of tool). Everything was about what I expected, until the blurb at the end, where they asked for my SAT scores.

I found this interesting for two reasons:

  • Since 2023 or so, I've noticed more and more tech companies and hyper-growth startups dropping bachelor's degree requirements. Instead, many ask about side projects/links to personal websites/blurbs about what you're most proud of. Basically, indicators that someone is passionate and a 'self-starter.'
  • Why would you ask for a self-reported, unverifiable test score that could be decades old at this point? There must be a better predictor for current cognitive ability than that, if that's what you're testing for. If not, this seems like a trick question to test compliance with pointless directives.

Under ideal conditions, SAT scores are probably a decent proxy for predicting whether a new hire will contribute to the success of your business. Cognitive ability is, unsurprisingly, positively correlated with professional success.

That said, there are clear problems with using an old SAT score as a decades-old litmus test for intelligence:

  • You're partly making your decision based on who someone was as a 17 year old.
  • Each applicant took the test under unknown conditions. You don't know if this was their first attempt or their fifth, whether they walked in cold or worked with a tutor for months, or whether they came from a perfectly stable home life or found out the night before that their dog died/parents are getting divorced/worked the night shift at a grocery store to help pay rent. Maybe they struggled with addiction and rebellion in their youth. There are probably many reasons why two candidates, equally cognitively capable, got different scores, or that a test taken a decade ago wouldn't reflect a candidate's current ability.

Regardless of whether they were testing compliance or cognitive ability, this unusual request led me down a rabbit hole of hiring practices. Both the gold-standard, time-tested methods and the bizarre, almost mystical attempts to see the future or know someone else's soul.

A very very brief history of hiring assessments

Historians credit the military with establishing the 'science' of the modern hiring assessment.

In 1917, staring down the fraught pitch of WWI, the army had to sort 1.5 million recruits quickly into the units for which they were best suited. Due to the time crunch, traditional interviews weren't gonna cut it.

Robert Yerkes and his committee developed the Army Alpha and Army Beta tests as a solve, the former for literate recruits and the latter for illiterate recruits or those who didn't speak English. They were basically the same, just Army Beta was nonverbal to circumvent communication barriers. These formed the basis of group-administered cognitive tests.

The other founding lineage is the assessment center, originally used for spy selection. Personality theorist Henry Murray ran a three day program at a bucolic estate tucked away in the country with simulations and multiple assessors. In the mid 1950s, AT&T launched a longitudinal study to see whether these assessment center ratings held water and found they worked decently well as a predictor for success.

The methods that appear to work best most often

No hiring method is perfect, of course, but we do have evidence that a mixture of a couple methods work best in predicting future professional success.

As of 1998, the cognitive test (of which SAT is one) was considered the best predictor. Then, in 2022, Sackett et al. argued convincingly that structured interviews, not cognitive assessments, were the strongest predictor of job performance.

As of the writing of this blog, these are the methods with the best evidence for predicting success on the job in knowledge work:

  • Structured interviews: These interviews must have consistent questions, a scoring rubric, and trained interviewers to work. They topped Sackett's list of 'best hiring assessments', but even they weren't super consistent (results had an 80% credibility interval). The TL;DR was that, naturally, the effectiveness of this method varies a lot based on how well-trained the interviewer is and how sound and consistent the rubric is.
  • Work sample tests and job knowledge tests: Take-home projects or a trial period of some kind. This makes the most intuitive sense by far: having candidates do a representative slice of the job gives you a solid idea of whether they'd be any good at it. Combining this with structured interviews was (before AI) considered a gold standard; you'd get a sense of who they are and how they work by talking, have a way to compare them pretty objectively to other candidates because of the structured and consistent nature of the interview process, and then you'd get a sense of how they apply their attributes practically to the job via the work exercise.
  • Cognitive ability tests: Still strong predictors, just not the strongest. Most useful for jobs with heavy training or learning demands, less useful for jobs where the demands closely match a candidate's prior experience.
  • Assessment centers (spy shit): Great for predicting managerial aptitude but extremely expensive and time consuming, obviously.

Methods that do not work

I ran into a couple that you could've guessed from a mile away wouldn't work (ex: Myers Briggs and handwriting analysis). One of the least effective predictors was unstructured interviews or 'chats', which was interesting, because this is the assessment I encounter most often.

I'm guilty of doing these myself. These are common because they're easy, obviously, but they're vulnerable to a slew of fallacies: impression management, inconsistent questioning, halo effects, and similar-to-me bias (I like you because you're me!).

An internal Google study of tens of thousands of interviews reportedly found little to no correlation between interviewer scores and eventual job performance.

I also stumbled upon a bunch of instances of bright, adventurous thinkers trying to brute force all the barriers to getting to know someone and their abilities by trying extreme outsider assessments.

One I thought was fun because people are crazy and I think that's great: in 1921, disappointed by a perceived lack of rigor among the latest crop of college graduates, Edison devised a 163 question trivia exam. Included a bunch of random questions he knew the answer to (one was 'what kind of wood are kerosene barrels made from', and the answer was not 'who cares?').

Edison also subjected candidates to the 'salt test'. He'd serve candidates soup, and if they salted it before tasting it, he'd allegedly disqualify them. His theory here was that this proved they operated on assumptions.

Unorthodox hiring assessments aren't solely confined to distant history. A bunch of tech companies have used them, too. Ex: I'm sure you've heard of the stupid brain teasers popularized by Microsoft and Google. They'd ask candidates questions like "why are manhole covers round?" Google's Head of HR, Laszlo Bock, called these brainteasers "a complete waste of time" that "don't predict anything" and serve only to "make the interviewer feel smart" in his book.

Then there was Zappos founder Tony Hsieh (smart, cool-seeming dude, tragic story). He had the airport screen and 'the offer'. The airport screen entailed Zappos flying a potential hire out. Upon arrival, they'd be picked up in a company shuttle. At the end of the day, the recruiter would ask the driver how they were treated. If they were treated badly, the candidate was not hired.

With 'the offer', after a 4 week time investment of training, each new hire would be offered $100 (and later up to $3,000) to quit on the spot. The goal was to see if they were really in it to win it or if they'd leave for easy money.

How to hire in the age of AI

Returning to why I'm writing this in the first place: I have no idea why this ETL startup asked for my SAT scores (have truly never seen that before in my adult life), but it is possible they're trying to get a read on raw, non AI-assisted intelligence.

It's definitely getting increasingly difficult to evaluate raw intelligence using the current gold standard methods post-AI. All resumes read the same and include every single keyword in the job description, and the cover letters all have the same 'hard-hitting, journalistic style'. It makes everyone seem like a perfect, phony all star. I can see why you'd default to a standardized test taken long before generative AI rose from the sand and slouched toward Silicon Valley to be born.

You could make the argument it doesn't matter, since the applicant can use AI at work, so whatever they send is still a fair representation of the quality of work they'd do on the job. But I get why you'd want to assess the raw horsepower of the pilot you'll have at the helm of even our most intelligent and autonomous tools.

I hope this one application isn't indicative of a larger trend that's coming our way. In case it is, here are a few alternatives I think would make more sense and that, as an applicant, I'd be totally willing to do:

  • Timed/live exercise (preferably in person if the role is in-office or hybrid. Otherwise, you could hire someone remotely or send a nearby employee to proctor the assessment).
  • You could still do a take home exercise, but then have the live meeting just be an interrogation about their work. If they're made to defend it or explain it, you'll probably have a better idea of whether the idea 'came from them', or if they at least understand what the AI generated enough to talk fluidly about it.
  • Bring back the in-person white boarding exercise.
  • For sales, live roleplay.

All of these exercises would have portions that need to be done face-to-face, either with a proctor or someone part of the hiring process. It's just too easy to use AI to take shortcuts if you're interfacing through a computer screen.

Even trying to hire people for my team in the last year, I've noticed many candidates' gaze darting to the side and staying there after a particularly meaty question. It looks a hell of a lot like they're reading and reciting a response from Claude, but, of course, I can't prove they're doing this over looking at their own notes or just getting a quick refresher on the job description. For those reasons, while it's certainly suspicious, I don't feel right taking points off for it.

I haven't kept up with the post-AI hiring discourse in tech so maybe these hiring practices are already in use. But it's clear the best predictors of future success tend to be assessments that are verifiable, current, and reasonably controlled. So we might have to return to the in-person proctored test with paper and pencil.

联系我们 contact @ memedata.com