119. If You Invite Me to Apply Unsolicited and Then Drop Me in an Early Round, I'm Going to Blacklist You

(Epistemic status: part prescription, part mechanism design and analysis run in cool anger, part price list, and part call to arms. If you recognize yourself on the other side of this, do better. If you can't recognize yourself on the other side of this, do WAY better. For anyone in AI safety who's spent weekend hours applying, called in favors for recommendations, and got their hopes up, and then received a form letter for their pains.)

It's a languid Tuesday evening when the email hits your inbox. "Invitation to Apply for an Upcoming Fellowship" is the topic. You've been trying for a grant, which so far hasn't gone as hoped, so you open it. You recognize the org's name; they do good work. You think their recent paper had some flaws, but on the whole their organizational heart seems to be in the right place, and they're doing the work.

"Hello!" it begins. Your name's in the next sentence. "We got your contact info from an alumni database from" - oh, right, that thing you did last summer - "and we think you'd be a strong fit for our program. We're scaling fast, and we'd love for you to apply." Maybe they even mention that paper you helped write a few months ago, or the mentorship you did for a few months for the one remote program. There's a link to the application, hosted on Airtable, and a deadline a couple of weeks into the future - tight, but you can probably manage if you shuffle some work around, set a project aside for a bit, and miss bingo night this week at the makerspace. "We're especially excited about people with your background in..." - what? Oh, right, you did write a sequence on that one topic a while back, didn't you? (It didn't get a lot of traction at the time, so it's nice that someone finally noticed.) Then there are some more details about the program itself. Maybe there's also the offer of an extended deadline, just for being someone they reached out to.

So you apply. Of course you do. This is the best opportunity you've had in weeks; you might put yourself back on the gameboard, maybe even end up with a position at a proper research group, or maybe you could use it a springboard for this one org you want to found. It'd mean dropping your backup plan, but it's a backup plan for good reason. You start to get a little excited. You want to do it right, so you set aside an evening to read everything the org has put out in the past year; you feed it all to an LLM as well, and then you feed in your notes, to see what you've missed or misunderstood. You fill in the demographic information for the application, cursing yourself when you accidentally refresh the page and it turns out that the org didn't turn on "save text in incomplete forms". You attach your resume - wait, no, you give it another look, realize that it's a few months out of date and also targeted towards your backup plan's industry, so you knock the dust off of it and reattach. You write the three-to-five short essays; you recognize one question from a few past applications and the rest as novel anti-inductive takes on other familiar questions; they're all of them variations on "why us, why you, why now, what makes you special, how culturally committed are you". You dig up the work sample that best fits what they claim to want and spend an evening knocking the dust off that, too. You warn two busy people that a reference request might be coming; then a third, when one of the first two nearly-politely declines. You thank the two that accept, and you strike the third from your list. Call it a couple of evenings and some weekend time if you're especially fast and efficient, and more like a week's worth of evenings and random free time if you're honest. You double-check everything. Then you hold your breath and click "Submit". And then you wait.

And then some time later - maybe it's a few days or a week and startles you with its speed, and maybe it's a couple of months and you've already moved on - you get another email. This one has your name mail-merged into the saluation, too, and nothing else about you anywhere in it. It's chilly; slickly corporate. You recognize the phrases here, too: "We received a large number of strong applications this round." "We've decided not to move forward with your candidacy at this time." "We encourage you to apply for future roles." "We regret that we cannot give individual feedback, but in general, unsuccessful applications had some commonalities..." You read the list. Literally none of these apply to you. You sigh; you un-star the invitation and the application receipt acknowledgement. You mentally strike-through a few to-do list items. You get up from your desk, get a drink, sip it slowly, rub your temples for a bit. You got dropped at the literal first cut - the résumé screen, before anyone could possibly have ruled you in. (Maybe you know because the posted next step was an interview; or because you have a friend on the inside; or because the recruiter admitted it when you ran into them at a party.) You wish you could find it in yourself to be more surprised. You wonder whether a human ever saw your application or if it was Sonnet that axed you; never mind whether a decision-maker did. (That was never in the cards.) You wonder why you keep doing this to yourself - why you keep getting your hopes up; why you still listen to the advice you got a few times to "just apply to everything relevant-looking". You wonder why this is how hiring works. Your phone buzzes in your pocket - an 80k Hours roundup email: did you know that there's this exciting new fellowship you should apply to? You recognize the name. You go to bed.

If you're about to jump to the comments to tell me "that's just how hiring works" or accuse me of being mid, kindly spare me. I know how hiring works, and if in bringing several original contributions, numerous posts and preprints, a deep and longtime personal pull towards AI safety, and a PhD in math to the table, I'm what counts as floor-sweepings, then I have no idea what any of these positions are looking for or how they hope to scale.

You know what we love doing here? Operationalizing wicked problems and gearsily analyzing why people do what they do. In that spirit, what really is that first email? What does it mean, and why does it exist? For one, it costs its sender approximately nothing: a template, a list scraped from a conference's roster or a fellowship's alumni database, a merge field for the name, and maybe a bell here or whistle there. It is thus direly low-entropy; it carries very little information about the recipient. It's top-of-funnel, in a field that claims to be hungry to scale and desperate for ready hands; put it that way and it almost sounds virtuous to ask hundreds or thousands of people to spend anywhere from 3 to 30 hours preparing. Put it that way, and you can almost forget that whoever sent it didn't and couldn't have evaluated you; you can almost forget that in borrowing the shape of a letter of introduction, it presents itself in the shape of something real and valuable to those who by construction are starving for just that shape of thing.

I've spilled ink on what happened to the letter of introduction in "108. Where Are All the Journeyman Researchers?", but as a refresher, a letter of introduction was a journeyman's travelling certificate. Once, it meant that its bearer was already competent, and that there was a specific and acclaimed person who would say as much for them, and that on the strength of the master's mark, the bearer should be hired. In our time, it has long since withered into the letter of recommendation, a piece of obligatory non-condemnation with no room left in it for actual signal. Negative room, actually, at this point: any letter lacking specific praise may as well be literally blank at best; so goes it, in this world where the Red Queen rules. The decay of the invitation to apply is the next stage of the same ratcheting decay: it uses the form of "we have looked at you and we specifically want to talk to you" to perform the function of "we have a number of applications to hit this quarter because scaling is inherently virtuous and no one can possibly be harmed by an offer". And the form is what you respond to! You'd have to be a fool not to, because in a sane world, that form means what it says. The whole trick to getting those application numbers up depends on your keeping your hopes up and living in the sane world for barely long enough to hit submit.

I've also written about chump checks in "94. Chump Checks" - the way a person or an institution passes or fails a quick test of its character and intent through its standard procedures: what it does to you as a matter of routine. This is one such. A procedure that solicits with a template, rejects with another one, routes little-to-nothing human between the two, and prides itself in casting as wide a net as possible is a procedure that has priced your time at zero negotiable American dollars.

It's not malicious - never malicious; maliciousness is a waste of time and no fellowship employs someone whose job it is to be malicious - but rather simply what falls out of the system of incentives as presented and felt. None of it requires villains, but none of it needs to. Let us sketch out the dance steps to waltz the primrose way to the everlasting bonfire. Keep triple time with me.

First, the inviter and the screener are different people, and they're scored on different things. The inviter gets scored on applications received - the more, the better and more virtuous. The screener has a different task: avoiding wasting the hiring manager's precious time. Neither of these people gets graded on how much aggregate time or effort or hope they waste, and even if they're the same person, they're told to wear two different hats. It's the second-oldest story in economics, after "gains from trade": negative externalities fall on outsiders, are invisible, and any effort spent preventing them is (from within) better spent literally anywhere else. Your weekend, to them both, is an acceptable loss.

Second, first-round rubrics screen for legibility, not fit. "Three years at a frontier lab"; a known advisor's name, or better yet, two or three; a degree with the right letters in the right order from a right place; a publication list that looks like other publication lists: these are the stuff of an application that has a better shot at a fellowship or a position. "Up to speed on The Problem" often doesn't survive a keyword filter; neither does "obviously one of us", which anyone who'd actually read your application would see in the first paragraph. The worst part is that the people the invitation flattered most - the illegible ones, the ones with five roofs in as many years and a trail of grant-funded work that "doesn't count" - are exactly the ones that the first cut is built to drop. The email, in effect, was written to stoke the hopes of precisely the kind of person the rubric was written to reject.

Third, the loss function is terribly lopsided - unavoidably so. An org dreams of a future star researcher and dreads missing that one unicorn vastly more than it worries about wasting a hundred solid role-players, and so it casts the widest net it can manage so that it can cut brutally and early; from inside, this is almost free. Only from inside, though: once again, the hundreds of weekends it burned don't show up as a line item. (And anyway, they've probably got it too good, they do. Getting rejected a few times builds character, gives them something to add to their list of failures to present proudly, as proof that they're striving. And there's not enough time left for gentleness.) No, they show up instead, months or years later, as "the mysteriously leaky talent pipeline", which that same org will write a thoughtful thinkpiece about, or mention alongside application statistics before touting their recent successful cohort and attractive upcoming run. Apply immediately!

That brings us to that infamous phrase: "talent-constrained". (And a spin and a dip and...) It was on everyone's lips a while back and crops up now and again, and it's still the harmony line we move to; easy to miss but ever-present, between the lilting bittersweet melody of research struggles and triumphs and the driving beat of will-we-manage-this-in-time-oh-shit-we-are-not-on-track-to. Our field is small because the people are scarce, not because the money is; so goes the coventional wisdom. Something about that sounds out of tune, though... if a field were truly talent-constrained, every application - let alone every solicited one! - would be read to the end by a human being, someone with any decision-making power and who fretted over missing a diamond in the rough. (Someone without a college degree, say, who was very smart and thought deeply and mostly just posted their weird but cogent takes on an obscure blog somewhere.) But instead, we turn away hopeful would-be attendees with a template. The constraint seems unlikely to be talent, then: it's - trust, vetting, the missing rung. Call it whatever you like: at its core, it's the inability to recognize a competent stranger at a cost that anyone's willing to pay - or put another way, an unwillingness to pay those costs at scale, or to scale to be able to pay them. Who it is mournfully intones that phrase should be another tip off: it's mostly the people who got through. From inside a pool filtered by networks, the pool looks shallow; that's what a filter does. Selection effects rule the day, as ever: the kind of thing you see in the wild has to have got there somehow, and the kind of person who says "we just can't find anyone" has, by construction, never been that "anyone".

It's not even just those darned emails, either. If it were, this would be a much shorter post. Those emails are just one visible symptom of a field that mints apprentices by the cohort - fellowships, programs, bootcamps, and so on, each with its own glossy alumni page, full of smiling faces - and then has nowhere to put them: a few labs and other outfits that hire from each other, a scattering of small orgs with a few seats apiece, and a howling nothing in between. Into that nothing fall people who are, by any test you'd care to name, obviously fits: they've read everything, they can talk about The Problem at any resolution you like, they've shipped work. They know their steps and their fingerings. They've danced and they've played and they've bowed, but now they'll be turned away at the door.

I've seen dozens of bright, talented, driven people put in this position over the last 4 years, and I've been in this position on and off, too: they're accepted to a fellowship, they produce good, interesting, novel work, they graduate, and then they have nowhere to go. Sometimes they're advised to found their own org, and so bounce off of grant orgs trying to scrape together funding to do it. Sometimes they limp along in a research group, unsure of whether they'll get another 3 months of funding. Sometimes they try to get by on a shoestring. In all these cases, it wears on them to keep trying, keep applying, keep offering, keep writing. So they leave: for academic posts, for a SWE or data science job, for a quant shop, for an academic post, or even for just plain being unemployed until they start bagging groceries or waiting tables instead of hoping and being repeatedly disappointed; anywhere where the recruiting process at least has the decency to be openly mercenary about it, and nobody pretends that a cold call is a social one. We are, collectively, complaining about a shortage of talent while turning away at the door those people that we ourselves wrote to - and this is the system working as intended! This is untenable - for applicants now, for the field and its leaders later, and for us all far sooner than anyone would like.

Accordingly, here's my policy going forward (and I invite you to join me): if you invite me to apply unsolicited to your thing and then you drop me in an early round, I'm going to blacklist you going forward. It's demonstrably just not worth it to me to spend time and effort and care and hope on your thing, and I'm not getting those back here, so I'm foreclosing on you and your org instead. I care about precision, and about saying exactly what I mean, so here's a few definitions, some operationalizations, a consequence, an repair, and some game-theoretically indicated largesse.

"Invited unsolicited" means outreach I didn't ask for and wasn't expecting; centrally by way of an alumni database, or a warm intro routed through a friend of the org. I'd call a specific person I'm friendly with writing to me to say that they personally want for me to apply - someone who knows my work and my name - "solicited"; newsletters and job boards don't count either way - broadcasts started out being openly for no one in particular.

"Dropped in an early round" means any rejection - especially a form rejection, with no sentence anyone wrote about me specifically - that arrives before I've had either of a synchronous conversation with a human empowered to advance me and an actual relevant technical screen. I'd give some number of days after the application deadline, too, but I've had orgs take months to finally tell me they'd actually rejected me weeks prior and had long since moved on to interviews.

"Blacklist" - the consequence - means that I will not apply to you again, invited or not, solicited or not, forever, by default. I'll give it some thought after three years. If anyone asks me why, I'll tell them, plainly, bluntly, and without adjectives or hedging, and I'll tell them roughly immediately in whatever room they asked in. No ill will, no campaign, no proactive name-and-shame, nothing like that - just a boundary and a price and someone who might have spoken well of you very much not doing that.

The repair is this: someone writes to me, in their own words, and says what happened. And I don't need or want an apology, I want an account. "We had over five hundred applications and a two-person admissions team, the screen was automated, and yeah, you probably shouldn't have been invited at all, given that you don't have a policy background." is a complete and sufficient sentence, and it cancels the blacklist on the spot. Or even: just publish your funnel! How many you invited, how many applied, how many applications were read by a human, how many you interviewed, how many you accepted, and how many accepted you back. Let the next person decide with their eyes open. I'd apply to an honest meatgrinder, but not one too ashamed or busy to be frank. One paragraph, one sentence of actual information: that'd be all it takes, that cheap of a fix, and so it's especially damning how rarely it comes.

The largesse is this: everyone gets one. Name changes don't reset it, and inevitably this is going to be partially retroactive, but every org still gets one. The first time, I will act as though I've assumed a mistake, because mistakes are common and overbooked oversights more so and malice is rare, and I'd rather be a fool once than a cynic immediately and forever. Zero chances would be unkind and unwise; infinite chances would make me a desperate doormat and less wise still; that just leaves one as the remaining number requiring no justification. The second time, you're on the list for good.

"Why a blacklist, and not just a grumble over drinks or a private shitlist to seed the whisper network with?" A reasonable question! My reasoning is, an individual applicant has precisely one lever to use: their future participation. But a lever of this sort that no one can see can move nothing. Make the refusal private, and it's nothing more than resentment, and that's a price you pay without charging anything on. That's why this rule has to be public, with legible terms, an exit clause, and a line drawn in advance: otherwise, you lock doors you should only close firmly, or alternatively when the next flattering email arrives - and arrive it will; you're on pool lists forever - it'll be all too easy to think "well, maybe this one, just this once". Judgement calls are hard and expensive. Bright lines are easier to hold.

"Wide nets catch the people who'd never have applied otherwise. Some of them are stars, and those stars justify the net. Can't you see that?" I can! Fine, then: say as much in the email. "This is a broad solicitation. Our first screen is automated. Expected time to a human read is X." I'll choose whether or not to apply with eyes open, and so will everyone else, and then your funnel will be smaller but more honest. I think you should want to make that trade. Honesty is pretty cheap, in the final analysis, looked at from all angles. Flattery, on the other hand, is what's been expensive, and you haven't noticed because you've normalized making the recipients pay for it.

"Isn't this petty?" It's a price, is what it is. Prices aren't petty; they're how anyone ever finds out what anything's worth. You've been getting evenings and weekends for free that the smart driven people who they belong to could have used for literally anything else, and you've been getting good or at least indifferent press from people who are too afraid to say anything else where a granter might hear it. That subsidy is ending, that's all.

"You'll miss out on a good role someday." Possibly! But a commitment that costs me nothing binds me to nothing. I've read my Schelling and (hopefully, plausibly, probably) so have you.

"But we're all on the same side." We are, or at least I should like to think so. (That's the sort of thing that's generally much more truthful when inferrable while unspoken.) That's why I'm telling you at nonzero personal cost and risk instead of just leaving. People who just leave don't write posts; they write LinkedIn updates about their exciting new chapter in data science for ads, or SWE for insurance adjustment, or the capabilities team at OAI, and you never find out what it cost you.

I don't like any of this, to be clear. I'd rather it make no sense to adopt this policy. I'd vastly prefer a world where an email with my name in it meant that someone had read my work and wanted more of it and thus wanted me, and where "we'd love for you to apply" was the first sentence of a conversation rather than the last sentence I'd get from a human. But (...and hesitation step, and chassé...) what's true is already so, and it's thus there to be interacted with, and interacting with this particular truth means acknowledging it and pricing it and setting a boundary.

Maybe you run hiring somewhere in this field and you've read this far with your jaw set: good. Rejoice, for you can cheaply do far better. If you mean what you say, then read the applications you ask for; read them all the way to the end, and do it yourself rather than feeding everything to an LLM. If you don't mean it, don't send it. Our field is small and its memory is long, and the people you're burning are the ones who'll be deciding whether to recommend you to the next cohort of exactly the people you say you can't find. I've already decided not to bother with three applications this past month, but delighted in two others.

And if you're the other sort of reader - the one with a folder of these emails and a matching set of uncategorized rejections, who's been lying awake wondering what's wrong with them - then I recommend taking my policy as your own, and take also this consolation, which rhymes with the previous one: it's not you. Your application was never going to be read. Take my price and boundaries and terms, or set your own. Hold them when the next flattering email comes, and value your weekends and your hope. I know all too well what they're worth.

Comments

Popular posts from this blog

82. Deadlock in the Parliament of the Self

98. The ICML 2026 Seoul Survival Guide

4. Seven-ish Words from My Thought-Language