Anthropic: "Applicants should not use AI assistants"
simonwillison.net
simonwillison.net
sshine reply above is coming from a very conflictual mindset. "Can I still use AI and not be caught? Is it cheating? Does it matter if it's cheating?"
I think that's a bit like lying on your first date. If you're looking to score, sure, it's somewhat unethical but it works. But if you're looking for a long term collaboration, _and_ you expect to be interviewed by several rounds of very smart people, then you're much better off just going along.
It’s not about being wrong, it’s about being ironic. We have LLMs shoved down our throats as this new way to communicate—we are encouraged to ask them to make our writing “friendlier” or “more professional”—and then one of the companies creating such a tool asks the very people most interested in it to not use it for the exact purpose we’ve been told it’s good at. They are asking you pretty please to not do the bad thing they allow and encourage everyone to do. They have no issue if you do it to others, but they don’t like it when it’s done to them. It is funny and hypocritical and pulls back the curtain a bit on these companies.
It reminded me of the time Roy Wood Jr visited a pro-gun rally where they argued guns make people safe, while simultaneously asking people to not carry guns because they were worried about safety. The cognitive dissonance is worth pointing out.
Sam Altman (2023): "something very strange about people writing bullet points, having ChatGPT expand it to a polite email, sending it, and the sender using ChatGPT to condense it into the key bullet points"
3 years ago people were poking fun about how restrictive the terms were - you could get your API key blocked if you used it to pretend to be a human. Eventually people just used other AIs for things like that, so they got rid of these restrictions that they couldn't enforce anyway.
https://x.com/sama/status/1631394688384270336
Thanks for this heads-up, by the way, I've missed this particular tweet, but eventually got into exact same observation.
Well, no. It's irony, but it's only cognitive dissonance in a comedy show which misses the nuance.
Most pro-gun organizations are heavily into gun safety. The message is that guns aren't unsafe if they're being used correctly. Most of the time, this means that most guns should be locked up in a safe, with ammo in a separate safe, except when being transported to a gun range, for hunting, or similar. When being used there, one should follow a specific set of procedures for keeping those activities safe as well.
It's a perfect analogy for the LLM here too. Anthropic encourages it for many uses, but not for the one textbox. Irony? Yes. Wrong? Probably not.
There’s probably actually some other hidden factor though, like the venue not allowing it.
Edit: FWIW those late night TV shows are nothing but rage bait low brow “comedy” that divides the country. But the above remains true.
That's not what the NRA is pushing for, any more than there are Democrats pushing for mandatory sex changes for all kids (yes, this is cited on similar right-wing comedy shows, and individuals on the right believe it). Pushing for a right doesn't mean 100% of the population will exercise that right.
And yes, most venues (as well as schools, government buildings, etc.) will not allow guns. If there's a security guard, police, or similar within spitting distance, there isn't a reasonable self-defence argument.
One of the interesting pieces of data is looking at 2nd amendment support versus distance to the nearest police station / police officer / likely law enforcement response times. It explains a lot about where support / opposition comes from.
> Everyone has a gun on their person with no licensing requirements. Yet at the NRA conference they ban guns.
yet, you claim that it's the late night TV that divides us, while making sure to double down on your misleading statement.
The NRA doesn't "ban guns at their conferences", they have been banned at small parts of a multi-day conferences e.g. where Trump was speaking because that was a rule established by the secret service and they complied for a small part of the conference.
When the majority of a conference allows guns, it's simply a lie to claim that guns were banned. An unintentional lie, I'm sure, but it seems likely to be the result of you believing some headline or tweet and accepting something wholesale as truth because it fit your narrative. I'm guilty of the same, it happens, but hopefully we can both get better about portraying easily fact checked things as the truth.
There is no irony.
Sure, some places in the world can successfully limit gun access. Those places aren't close and even bordered by the most active cartels in the world.
Just as a fun thought exercise, consider that to grow the plant necessary to produce just a single drug, cocaine, for the country, every year. It takes at least 300,000 acres, or roughly the size of los angeles. That's after decades of optimizations to reduce the amount of land needed. It's also only for one drug among a vast number that are regularly consumed in the US.
In relation, you can 3d print guns at home. Successful builds have been made from some of the cheapest 3d printers you can find.
When have criminals EVER followed law, code, rules, or even a suggestion from their fellow citizens?
Believing laws deter criminals is almost criminally insane and beggars all logic.
After all of the accumulated evidence against your belief, you still believe laws deter criminality.
The death penalty doesn’t deter criminals. How could words possibly have an effect?
I have no illusion that safety or education is an actual concern in public schools in general.
Like the nuance between sending out your love and doing the Nazi salute? Or different?
* Most of the AI you get shoved down your throat is by the providers of services you use, not the AI companies.
* Among the AI companies, Anthropic in particular has had a very balanced voice that doesn't push for using AI where it doesn't belong. Their marketing page can barely be called that [0]. Their Claude-specific page doesn't mention using it for writing at all [1].
You seem to be committing the common fallacy of treating a large and disparate group of people and organizations as a monolith and ascribing cognitive dissonance where what you're actually seeing is diversity of opinion.
Being able to gauge a candidate's natural communication skills is highly useful. If you're an ineffective communicator, there's a good chance your comprehension skills are also deficient.
IMHO, an effective interview process should attempt to mimic the position for which a person is applying. Making a candidate jump through hoops is a bit disrespectful.
We are quickly moving into a world where most communications are at best assisted by AI and more often have little human input at all. There’s nothing inherently “wrong” about that (we all have opinions there), but “natural” (quotes to emphasize that they’re taught and not natural anyway) communication skills are going to be less and less common as time marches on, much like handwriting, typewriting, calligraphy, etc.
It'd have been ironic if Anthropic had asked the applicant not to use AI for the sake of originality and authenticity; but if the applicant felt compelled to do so, then it better rock and wow them to hire the applicant sight unseen.
It's sardonic because Anthropic is implying use of AI on them is an indication of fraud, deceit, or incompetence; but it's a form of efficiency, productivity or cleverness when used by them on the job!
Doesn’t seem like that to me, after reading the dictionary definitions.
Sardonic: grimly mocking or cynical.
Ironic: happening in a way contrary to what is expected, and typically causing wry amusement because of this.
Pretty sure I meant the second one.
I think they will get better results by having applicants talk to an AI during the application process.
I don't think that's true at all.
The writing was on the wall that the main use will be spam and phishing.
You can say the creators did not intend on this purpose, but it was created with knowledge that this would be the main use case.
I have read that people are getting more selfish[1], but it still shocks me how much people are willing to push individualism and selfishness under the guise of either, "Well it's not illegal" or "Well, it's not detectable".
I think I'm just very much out of tune with the zeitgeist, because I can't imagine not going along with what's a polite request not to use AI.
I guess that puts me at a serious disadvantage in the job market, but I am okay with that, I've always been okay with that. 20 years ago my cohort were doing what I thought were selfish things to get ahead, and I'm fine with not doing those things and ending up on a different lesser trajectory.
But that doesn't mean I won't also air my dissatisfaction with just how much people seem to justify selfishness, or don't even regard it as selfish to ignore this request.
[1] https://fortune.com/2024/03/12/age-of-selfishness-sick-singl...
No, what you are is ignoring the context.
This request comes from a company building, promoting, and selling the very thing they are asking you not to use.
Yes, asking you not to use AI is indeed a polite request. It is one you should respect. “The zeitgeist” has as much people in favour of AI as against it, and picking either camp doesn’t make anyone special. Either stance is bound to be detrimental in some companies and positive in others.
But none of that matters, what makes this relevant is the context of who’s asking.
Are you implying that Anthropic specifically pushes for their models to be used inappropriately as long as they're not the victims of that inappropriate use? Because I haven't seen that at all with Anthropic, they've been consistently the most subdued and reserved AI company out there, barely marketing their products at all and when they do, doing so very carefully.
Your reactions in this thread are understandable as reactions against the oversaturation of AI, but it's not really fair to paint all of the companies with the same brush when Anthropic exists to be a foil to Altman's irresponsible push for saturation.
Have you seen the job market? Companies will treat you like garbage through the interview process, make you jump through pointless hoops, and then even if you get the job you can be laid off at any moment because of arbitrary reasons the CEO made up to get their bonus.
Why should anyone be honest when it goes entirely unappreciated and unrewarded? I can completely understand why people would cheat. When companies stop treating workers like garbage then they deserve honesty.
This is a good example of the attitude that I'm describing.
Your question is close to an unthinkable culture shock to me.
In my values and ethics, honesty isn't transactional. It's not something you practice because you expect the same back. It's not something that you regulate and only provide to others that meet some moral bar that you set.
Honesty is just something you do because it's ethically right to do so.
( Nor by the way, is it motivated out of some fear of omnipotent reprisal. )
I think you're more accurate here, fuck them. The reality is I'm fortunate to be an office drone instead of treated as utterly disposable in a gig economy. And if someday AI gets good enough to replace me, I will be replaced.
What's good for the gander.... I promise you they will use AI to vet your application.
So?
I honestly fail to see even the irony. "Company that makes hammers doesn't want you to use hammers all the time". It's a tool.
But if I squint, I _can_ see a mean-spirited "haha, look at those hypocrites" coming from people who enjoy tearing others down for no particular reason.
Even when squinting I can't see a genuine argument for why Anthrpoic shouldn't be raked over the coals for their sheer hypocrisy
The optics are just bad. Stand behind your product, or accept that you will be fighting ridicule and suspicion endlessly.
In a sense, they poisoned the well and don't want to drink from it now. Looking at it from this perspective justifies (in the eyes of some people at least) said cheating. Something something catalytic converter from the company truck.
Sorry, the thought process of thinking that using an LLM for a job application, esp. for a field which requests candid input about one's motivation is acceptable, is beyond me.
Of course, we're in a free world. People are free to do what they wish, and face the consequences, of course.
This is just experience and seniority in general, nothing particular about LLMs. For most businesses, I would behave the same.
Observation/Implication/Opinion:
Think reciprocal forces and trash TV ethics in both, closed and open systems. The consequences are continuously diminished AND unvarying returns. Professionally as well as personally, in all parties involved. Stimulating, inspiring, motivating factors as well as the ability to perceive and "sense" all degrade. But compensation and cheating continue to work, even though, the quality of the game, the players and their output decreases.
Nothing and nobody is resilient "enough" to the mechanism force - 'counter'-force so you better pick the right strategy. Waiting/Processing for a couple of days, lessons and honest attempts yields exponentially better results than cheating.
Companies should beware of this if they expect results that are qualitatively AND "honestly" safe & sound. This has been ignored in the past decades, which is why we are "here". Too much work, too many jobs, and way too many enabling outs have been lost almost irreversibly, on the individual level as well as in nano-, micro-, and macro-economics.
Applicants using AI is fine but applicants not being able to make that output usefully THEIRS is a problem.
First off: Since your opinion is the popular one, "sshine reply below". ;-)
Those are not questions I ask.
Using AI is never cheating, you're optimizing for the wrong thing. Cheating occurs in games.
You can use AI, but if your writing ability is so far from what's expected, AI will just make that obvious (sludge). If it's not far off, and you use it as one might use spellcheck or Grammarly, it's like wearing fake heels at an audition: To get noticed. Just don't come on stilts.
I imagine they wouldn't have a problem with using AI to proofread your responses. That feels like fair game.
But... it can't tell you your own thoughts. It has no thoughts. And if it did they certainly aren't your thoughts.
https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...
There are two backwards things with this:
1) You can't ask people to not use AI when careful, responsible use is undetectable.
It just isn't a realistic request. You'll have great replies without AI use and great replies with AI use, and you won't be able to tell whether a great reply used AI or not. You will just be able to filter sludge and dyslexia.
2) This is still the "AI is cheating" approach, and I had hoped Anthropic to be thought leaders on responsible AI use:
In life there is no cheating. You're just optimizing for the wrong thing. AI made your homework? Guess what, the homework is a proxy for your talent, and it didn't make your talent.
If AI is making your final product and you're none the wiser, it didn't really help you, it just made you addicted to it.
Teach a man to fish...
I'm an experienced dev (20 years of C++ and plenty of other stuff), and I frequently work with younger students in a mentor role, e.g. I've done Google Summer of Code three times as a mentor, and am also in KDE's own mentorship program.
In 2023/24, when ChatGPT was looming large, I took on a student who was of course attempting to use AI to learn and who was enjoying many of the obvious benefits - availability, tailoring information to his inquiry, etc. So we cut a deal: We'd use the same ChatGPT account and I could keep an eye on his interactions with the system, so I could help him when the AI went off the rails and was steering him into the wrong direction.
He initially made fast progress on the project I was helping him with, and was able to put more working code in place than others in the same phase. But then he hit a plateau really hard soon after, because he was running into bugs and issues he couldn't get solutions from the AI for and he just wasn't able to connect the dots himself.
He'd almost get there, but would sometimes forget to remove random single lines doing the wrong thing, etc. His mental map of the code was poor, because he hadn't written it himself in that oldschool "every line a hard-fought battle" style that really makes you understand why and how something works and how it connects to problems you're solving.
As a result he'd get frustrated and had bouts of absenteeism next, because there wasn't any string of rewards and little victories there but just listless poking in the mud.
To his credit, he eventually realized leaning on ChatGPT was holding him back mentally and he tried to take things slower and go back to API docs and slowly building up his codebase by himself.
I totally agree with this and I really like that way of wording it.
You are directly loaded with all the shiny tools and, while it does make it interesting and fun at first, the magic wears off rather quickly.
On the other hand, when you had to fight and learn your way up to level 80, you have this deeper and well-earned understanding of the game that makes for a fantastic experience.
Absolutely true. However:
The real value of AI will be to *be aware* when at that local optimum, and then - if unable to find a way forward - at least reliably notify the user that that is indeed the case.
Bottom line, the number of engineering “hard thought battles” is finite, and should be chosen very wisely.
The performance multiplier that LLM agents brought changed the world. At least as the consumer web did in the 90s, and there will be no turning back.
This is like a computer company around 1980, would be hiring engineers but forbade access to computers for some numerical task.
Funny, it reminds me the reason Konami MSX1 games look like they do, compared to the most of the competition: having access to superior development tools - their HP hardware emulator workstations.
If you are unable to come up with a filter for your applicants that is able to detect your own product, maybe you should evolve. What about asking an AI how to solve this? ;)
So as a mentor, you totally talked directly with them about what excites them, tied it to their work, encouraged them to talk about their frustrations openly, helped them develop resilience by showing them towards setbacks are part of the process, and helped give them a sense of purpose and see how their work contributes to a bigger picture, to directly address the side effects of being a human with emotions which could have happened regardless of the tool they used, and didn't just let them flounder because of your personal feelings about a particular tool they used, right? Or do you only mentor winners, and you've never had a mentee hit a wall before LLMs were invented and never had to help anyone through some of the impacts from emotional lows that an immature intern might need help from a mentor to work through.
If US can be the worlds biggest economy while having an opiod epidemy and writing paper cheques, if Germany can be Europes manufacturing hub while using faxes, sure we as a society can live in the unoptimal state of everything digital being broken 10% of the time insteaf of hald percent
I have a rule for myself as a non-native English speaker: Any day I ask LLMs to fix my English, I must read 10 pages from traditionally published books (preferably pre-2023). Just to prevent LLM from dominating my language comprehension.
Sometimes it is more important to get a point across in another language than it is to learn that language. Computers being automatable, you can use it to create a backlog for when you skipped learning so that you can maintain some control of your habit of not learning what you're saying.
I wonder if the same thing will happen with coding and LLMs.
With sat nav I don't even try to read the exit signs; I just follow the blue line. It takes me 10-20 drives somewhere before I have the muscle memory, and I never made an active mental effort.
Going somewhere by public transportation or foot, e.g. a large homogenic parking lot complex, I consciously make an effort to take mental pictures so I can backtrack or traverse perfectly the second time; in spite of that being mentally challenging, it's still the easiest way I have.
I cannot assemble the hardware that I write code for. This is in spite of having access to both the soldering equipment, the parts and the colleagues who are willing to help me.
At some point all skills become abstract; efficiency is traded for flexibility when you keep doing the same thing for a very long time.
I can still drive a stick shift, but maybe not in 20 years.
In many ways people that don't use sat nav are at a disadvantage: real time traffic and redirection, high precision ETA, trip logging, etc.
That's a scary thing
Sometimes I have qwen2.5-coder:14b whip up a script to do some little thing where I don't want to spend a week doing remedial go/python just to get back to learning how to write boilerplate. All that experience means I can edit it easily enough because recognition kicks in and drags the memory kicking and screaming back into the front.
I quickly discovered it was essentially defaulting to "absolute novice." No error handlers, no file/folder existence checking, etc. I had to learn to put all that into the prompt.
>> "Write a python script to scrape all linked files of a certain file extension on a web page under the same domain as the page. Follow best practices. Handle errors, make strings OS-independent, etc. Be persnickety. Be pythonic."
Here's the output: https://gist.github.com/kyefox/d42471893de670a2a4179482d3c8b...
I'm far from an expert and my memory might be foggy, but that looks like a solid script. I can see someone with less practice doing battle with debuggers trying the first thing that comes out without all the extra prompting hitting errors and not having any clue.
For example: I wrote a thing that pulled a bunch of JSON blobs from an API. Fixing the "out of handles" error is how I learned about file system and network default limits on open files and connections, and buffering. Hitting stuff like that over and over was educational and instilled good habits.
It is a paradigm shift, yes. And you will know less about the implementation at times, yes. But will you care when you can deploy things twice, three times, five times as fast as the person not using AI? No. And also, when you want to learn more about a specific bit of the AI written code, you can simply delve deep into it by asking the AI questions.
The AI right now may not be perfect, so yes you still need to know how to code. But in 5 years from now? Chances are you will go in your favorite app builder, state what you want, tweak what you get and you will get the product that you want, with maybe one dev making sure every once in a while that you’re not messing things up - maybe. So will new devs need to know high level programming languages? Possibly, but maybe not.
2. Compilers are deterministic. You can recompile the source code and get the same assembly a million times.
You can also take a bit of assembly then look at the source code of the compiler and tell exactly where that assembly came from. And you can change the compiler to change that output.
3. Source code is written in a formal unambiguous language.
I’m sure LLMs will be great at spitting out green field apps, but unless they evolve to honest to goodness AGI, this won’t get far beyond existing low code solutions.
No one has solved or even proposed a solution for any of these issues beyond “the AI will advance sufficiently that humans won’t need to look at the code ever. They’ll never need to interact with it in any way other than through the AI”.
But to get to that point will require AGI and the AI won’t need input from humans at all, it won’t need a manager telling it what to build.
The point of coding is to remove ambiguity from the specs.
"Code" is unambiguous, deterministic and testable language -- something no human language is (or wants to be).
LLMs today make many implementation mistakes where they confuse one system with another, assume some SQL commands are available in a given SQL engine when they aren't, etc. It's possible that these mistakes will be reduced to almost zero in the future.
But there is a whole other class of mistakes that cannot be solved by code generation -- even less so if there's nobody left capable of reading the generated code. It's when the LLM misunderstands the question, and/or when the requirements aren't even clear in the head of the person writing the question.
I sometimes try to use LLMs like this: I state a problem, a proposed approach, and ask the LLM to shoot holes in the solution. For now, they all fail miserably at this. They recite "corner cases" that don't have much or anything to do with the problem.
Only coding the happy path is a recipe for unsolvable bugs and eventually, catastrophe.
Personally I see it more like going to someone who (claims) to know what they're doing and asking them to do it for me. I might be able to watch them at work and maybe get a very general idea of what they're doing but will I actually learn something? I don't think so.
Now, we may point to the fact that previous generations railed at the degeneration of youth through things like pocket calculators or mobile phones but I think there is a massive difference between these things and so-called AI. Where those things were tools obligatorily (if you give a calculator to someone who doesn't know any formulae it will be useless to them), I think so-called AI can just jump straight to giving you the answer.
I personally believe that there are necessary steps that must be passed through to really obtain knowledge and I don't think so-called AI takes you through those steps. I think it will result in a generation of people with markedly fewer and shallower skills than the generations that came before.
The use of AI is not just a labour saving device, it allows the user to bypass thinking and learning. It robs the user of an opportunity to grow. If you don't have the experience to know better it may be able to masquerade as a teacher and a problem solver, but beyond a trivial level relying on it is actively harmful to one's education. At some point the user will encounter a problem that has no existing answer in the AI's training dataset, and come to realise they have no real foundation to rely on.
Code generative AI, as it currently exists, is a poisoned chalice.
"We shape our tools and our tools in turn shape us" said Marshall McLuhan.
Problem with AI is that it is often black box tool. And not even deterministic one.
It’s fine to do in some cases, but it certainly gets abused by lazy incurious people.
Tool use in general certainly can be lazy. A car is a tool, but most people would call an able bodied person driving their car to the end of the driveway to get the mail lazy.
So I suppose you think that becoming lazy is always irresponsible?
It seems to me, then, that either the Amish are right, or there is a gray zone.
Being a CS teacher, my use of "responsible AI use" probably comes from a place of need: If I can say there is responsible AI use, I can pull the brake maybe a little bit for learners. It seems like LLMs in all their versatility are a great disservice to students. I'm not convinced it's entirely bad, but it is overwhelmingly bad for weak learners.
I showed this to my new colleague who is a bit older than me and sort of had similar attitudes as you. He told me he can do the same with some multi cursor shenanigans and I'll be honest in that I wasn't interested in his approach. Seems like he would've taken more time to solve the same problem even though he had superior technique than me. He said sure it takes longer but I need to verify by reading the whole class list and that's a pain but I just reloaded the page and it was fine. He still wasn't comfortable with me using copilot.
So yes, it does make me lazier but you could say the same about using go instead of C or any higher level abstraction. These tools will only get better and more correct. It's our job to figure out where it is appropriate to use them and where it isn't. Going to either extremes is where the issue is
When you take all three in consideration, an llm won’t really matter unless you don’t know much about the language or the libraries. When people goes on about Vim or Emacs, it’s just that it makes the whole thing go faster.
And you must use that brain muscles otherwise your skills became to degrade fast, like really fast.
As long as you ask llm What - or high level How - you should be good.
As soon as you ask for (more than trivial) code or solutions - you start losing your skill and value as a developer.
The argument that AI is bad and anyone who uses it ends up in a tangled mess is only your perspective and your experience. I’m way more productive using AI to help me than I ever was before. Yes, I proofread the result. Yes, I can discern a good response from a bad one.
AI isn’t a replacement for knowing how to code, but it can be an extremely valuable teacher to those orgs that lack proper training.
Any company that has the position that AI is bad, and lacks proper training and incentives for those that want to learn new skills, isn’t a company I ever want to work for.
Yeah, and I'd like to emphasize that this is qualitatively different from older gripes such as "calculators make kids lazy in math."
This is because LLMs' have an amazing ability to dream up responses stuffed with traditional signals of truthfulness, care, engagement, honesty etc... but that ability is not matched by their chances of dreaming up answers and ideas that are logically true.
This gap is inevitable from their current design, and it means users are given signals that it's safe for their brains to think-less-hard (skepticism, critical analysis) about what's being returned at the same moments when they need to use their minds the most.
That's new. A calculator doesn't flatter you or pretend to be a wise professor with a big vocabulary listening very closely to your problems.
This trope is unbecoming of anyone sensible.
Google 'reflections on trusting trust'. Your level of trust in software that purports to think for you out of a multi-gig stew of word associations is pretty intense, but I wouldn't call it pretty sensible.
I have the idea that there are 2 kinds of people, those avidly against AI because it makes mistakes (it sure does) and makes one lazy and all other kinds of negative things, and those that experiment and find a place for it but aren't that vocal about it.
Sure you can go too far, I've heard someone in Quality Control Proclaim "ChatGPT just knows everything, its saves me so much time!" To which I asked if they heart about hallucinations and they hadn't, they'd just been copying whatever it said into their reports. Which is certainly problematic.
At least in theory that’s not what homework is. Homework should be exercises to allow practicing whatever technique you’re trying to learn, because most people learn best by repeatedly doing a thing rather than reading a few chapters of a book. By applying an LLM to the problem you’re just practicing how to use an LLM, which may be useful in its own right, but will turn you into a one trick pony who’s left unable to do anything they can’t use an LLM for.
That seems like responsible use.
I can't really see an optimistic long-term result from that, similar to giving kids an iPad at a young age to get them out of your hair: shockingly poor literacy, difficulty with problem solving or critical thinking, exacerbating the problems with poor attention span that 'content creators' who target kids capitalise on, etc.
I'm not really a fan of the concept of homework in general but I don't think that swapping brain power with an OpenAI subscription is the way to go there.
(I still mentally cringe as I remember the posts about Disney and Marvel going out of business because of Stable Diffusion. That certainly didn't age well.)
I'm not saying they are essentially bad, or that they are not useful at all, far from that. But it's about the use they are given.
> You can use an LLM for everything, including curing cancer and fixing climate change.
Maybe, yes. But the danger is rather in all the things you no longer feel you have a need to do, like learning a language, or how to properly write, or read.
LLM for everything is like the fast-food of information. Cheap, unhealthy, and sometimes addicting.
Well, no. Homework is an aid to learning and LLM output is a shortcut for doing the thinking and typing yourself.
Copy and pasting some ChatGPT slop into your GCSE CS assignment (as I caught my 14yo doing last night…) isn’t learning (he hadn’t even read it) - it’s just chucking some text that might be passable at the examiner to see if you can get away with it.
Likewise, recruitment is a numbers game for under qualified applicants. Using the same shortcuts to increase the number of jobs you apply for will ultimately “pay off” but you’re only getting a short term advantage. You still haven’t really got the chops to do the job.
Thankfully it's usually pretty easy to spot this so it's basically an immediate rejection.
Too bad for them, but works for you…
I'm imagining a hiring workflow, for a role that is not 'specifically use AI for a thing', in which there is no suggestion that you shouldn't use AI systems for any part of the interview. It's just that it's an auto-fail, and if someone doesn't bother to hide it it's 'thanks for your time, bye!'.
And if they work to hide it, you know they're dishonest, also an auto-fail.
Adding a load of pseudo-science to the already horrible process of looking for a job is definitely not what we need.
I'll never submit myself to pseudo-IQ tests and word association questions for a job that will 99.9% of the time ask you to build CRUD applications.
The lengths that companies go to avoid doing a proper job at hiring people (one of the most important jobs they need to do) with automatic screening and these types of interviews is astonishing.
Good on whoever uses AI for that kind of shit. They want bullshit so why not use the best bullshit generators of our time?
You want to talk to me and get a feeling about how I'd behave? That's totally normal and expected. But if you want to get a bunch of written text to then run sentiment analysis on it and get a score on how "good" it is? Screw that.
Assume their command of English is insufficient to get the job ultimately. They've just wasted their own time and the company's time in that situation.
I imagine Anthropic is not short of applicants...
We're close to the point where using a human -> stt -> llm -> tts -> human pipeline you can do real time high quality bi directional spoken translation on a desktop.
I’ve read multiple LLM job applications, and every single time I’d rather have just read the prompt. It’d be a quarter of the length and contain no less information.
I think homework is more than a guide to what you should review though. It's partly so that the teacher can find out what students have learned/understood so they can adapt their teaching appropriately. It's also because using class/contact time to do work that can be done independently isn't always the best use of that time (at least once students are willing and capable of doing that work independently).
I think that's wishful thinking. You're underestimating how much people can tell about other people with the smallest amount of information. Humans are highly attuned to social interactions and synthetic responses are more obvious that you think.
At that time just downgrading them was justifiable, because though they had copied in someone else's text, they often weren't competent to identify the text that was best to copy, and they had to write some of the text themselves to make it appear a coherent whole and they weren't competent to do that. If they had used LLMs we would have been stuck. We would be sure they had cheated but their essay would still be better than that of many/most of their honest peers who had tried to demonstrate relevant skill and knowledge.
I think there is no solution except to stop assigning essays. Writing long form text will be a boutique skill like flint knapping, harvesting wild tubers, and casting bronze swords. (Who knows, the way things are going these skills might be relevant again all too soon.)
There's a difference between knowing how to use a calculator and knowing how to do math. The same goes for AI. Being talented at giving AI prompts doesn't mean you are generally talented or have AI-unrelated talents desired by an employer.
You can't make a rule, if people can cheat undetectably?
No, the homework is a proxy for measuring what the homework provider is interested to evaluate (hopefully). Talent is a vague word, and what we can consider talent worth nurturing might be consider worthless from some other perspective.
For example, most schools will happily give you many nation myths to learn and evaluate how much of it you can restitute on demand. They will far less likely ask you to provide you some critics of these myths, to search who created them, with which intentions and which actual effects on people at large scale from different perspective and metrics available out there.
Huh?
>You're just optimizing for the wrong thing. AI made your homework? Guess what, the homework is a proxy for your talent, and it didn't make your talent.
Isn't that the definition of cheating? Presenting a false level of talent you don't possess?
This really rubs me the wrong way because it reflects the shallow, borderline sociopathic stance of nowadays entrepreneurship.
Obedience to rules and honest attitudes is more than just an annoyance in your way to get rich, its the foundation of cooperation -- our civilization.
AI just lets you get deeper down the fake-it-until-you-make-it hole. At some point it will actually matter that you know how to do something, and good luck then.
Either for you, or for your customers, or both.
Oh, my sweet summer child
If I ask people not to use AI on a task where using AI is advantageous and undetectable then I'm going to discriminate against honest people.
They want to use AI in their hiring process. They want to be able to offload their work and biases to the machine. They just don't want other people to do it.
There's a reason that the EU AI legislation made AI that is used to hire people one of the focal points for action.
It might even give the honest people an advantage by giving them a tip to answer on their own.
LLMs are essentially translation tools. I use them to translate my picture-thinking into words, just like others might use spell-checkers or dictation software. They don't change my ideas or insights - they just help me express them in a neurotypical-friendly format.
The irony here is that Anthropic is developing AI systems supposedly to benefit humanity, yet their application process explicitly excludes people who use AI as an accessibility tool. It's like telling someone they can't use their usual assistive tools during an application process.
When they say they want to evaluate "non-AI-assisted communication skills," they're essentially saying they want to evaluate my ability to communicate without my accessibility tools. For me, AI-assisted communication is actually a more authentic representation of my thoughts. It's not about gaining an unfair advantage - it's about leveling the playing field so my ideas can be understood by others.
This seems particularly short-sighted for a company developing AI systems. Shouldn't they want diverse perspectives, including from neurodivergent individuals who might have unique insights into how AI can genuinely help people think and communicate differently?
I think it was the spell-checker analogy that really sold me. And this ties in with the whole point that "AI" isn't one thing, it's a huge spectrum. I really don't think there's anything wrong with an interviewee using an editor that highlights their syntax, for example.
Where do you draw the line, though? Maybe you just don't. You conduct the interview and, if practical coding is a part of it, you observe the candidate using AI (or not) and assess them accordingly. If they just behave like a dumb proxy, they don't get the job. Beyond that, judge how dependant they are on AI and how well they can use it as a tool. Not easy, but probably better than just outright banning AI.
It's interesting that most software departments now expect their staff to use AI tools day-to-day, yet many still ban it during interviews. Why not flip this around? Let candidates demonstrate how they actually work with AI. It would be far more valuable to assess someone's judgment and skill in using these tools rather than pretending they don't exist.
If a candidate wants to show how they leverage AI in their workflow, that should be seen as a positive - it demonstrates transparency and real-world problem-solving approaches. After all, you're hiring someone for how they'll actually work, not how they perform in an artificial AI-free environment that doesn't reflect reality.
The key isn't whether someone uses AI, but how effectively they use it as part of their broader skillset. That's what companies should really be evaluating.
I regard it as similar to using a text-to-speech tool for a blind person - who cares how they get their work done? I care about the quality of their work and my ability to interact with them, regardless of the method they use to get there.
Another example I would give: imagine there's someone who only works as a pair programmer with their associate. Apart, they are completely useless. Together, they're approximately 150% as productive as any two programmers pairing together. Would you hire them? How much would you pay them as a pair? I submit the right answer is yes, and something north of one full salary split in two. But for bureaucracy I'd love to try it.
This is exceptional writing and I appreciate your insight. Thanks for proving me wrong and giving me the chance to get closer to being right.
We don’t want to work with AI, we are going to pay the person for the persons time, and we want to employ someone who isn’t switching off half their cognition when a hard problem approaches.
And ultimately, this is what this is about, right? Delivering working products.
"That works" is doing a lot of heavy lifting here, and really depends more on the technical skills of the person. Because, shocker, AI doesn't magically make you good and isn't good itself.
Anyone can prompt an AI for answers, it takes skill and knowledge to use those answers in something that works. By prompting AI for simple questions you don't train your skill/knowledge to answer the question yourself. Put simply, using AI makes you worse at your job - precisely when you need to be better.
I don't follow.
Usually jobs require deliver working things. The more efficient the worker knows his tools(like AI), the more he will deliver -> the better he is at his job.
If he cannot deliver reliable working things, because he does not understand the LLM output, then he fails at delivering.
You absolutely can be a great developer who can't use AI effectively, or a mediocre developer who is very good with AI.
That's not the assumption. The assumption is that if you prove you have a firm grip on delivering things that work without using AI, then you can also do it with AI.
And that it's easier to test you when you're working by yourself.
If people can AI their way into the position you are advertising, then at least one of the following two things have to be true:
1) the job you are advertising can be _literally_ solved by AI
2) you are not tailoring your interview process properly to the actual job that the candidate will need to do, hence the handwave-y "oh well harder problems will come up later that the AI will not be able to do". Focus the interview on the actual job that the AI can't do, and your worries will disappear.
My impression is that the people who are crying about AI use in interviews are the same people who refuse to make an effort themselves. This is just the variation of the meme where you are asked to flip a red black tree on a whiteboard, but then you get the job, and your task is to center a button with CSS. Make an effort and focus your interview on the actual job, and if you are still worried people will AI their way into it, then what position are you even advertising? Either use the AI to solve the problem then, or admit that the AI can't solve this and stop worrying about people using it.
When we’re hiring for my role, Security Operations, I can’t have someone googling or asking AI what to do during an cyber security incident, but they can certainly use AI as much as they want when writing automations.
I reject candidates at all stages for all sorts of reasons, but more and more candidates believe the job can be done with AI. If we wanted AI, we will probably go wholesale and not include the person asking for the job to do the typing for us.
We’re not crying due to AI, we’re crying over the dozens of lost hours of interviews we’re having to conduct where it’s business critical that people know their stuff — engineering positions with consequences (banks, infrastructure, automotive). There isn’t space for “well I didn’t write the code”.
If your interview problems are representative of the work that you actually do, and an AI can do it as well as a qualified candidate, then that means that eventually you'll be out-competed by a competitor that does want to work with AI, because it's much cheaper to hire an AI. If an AI could do great at your interview problems but still suck at the job, that means your interview questions aren't very good/representative.
Where the line is drawn is context dependent, drawing the same single line for all possible situations is not possible and it's stupid to do so.
Sure, maybe you can say that the tasks should be complex enough that AI can't do it, but AI systems are constantly changing, collecting user prompts and training to improve on them. And sometimes the candidates aren't deep enough in the hiring process yet to justify spending significant time giving a complex task. It's just easier and more effective to just say no AI please
That did happen and the result from the test creators was the same back then: "we're not the problem, the machines are the problem. ban them!"
In the long run it turned out that if you could cheat with a calculator though, it was just a bad test....
I think there is an unwillingness to admit that there is a skill issue here with the test creators and that if they got better at their job they wouldnt need to ban candidates from using AI.
It's surprising to hear this from anthropic though.
I believe it won't change the outcomes that much. For example, on coding, an AI can't teach someone to program or reason in the spot, and the purpose of the interview never was to just answer the coding puzzle anyway.
To me it's always been about how someone reasons, how someone communicates, people understanding the foundations (data structure theory, how things scale, etc). If I give you a puzzle and you paste the most optimized answer with no reasoning or comment you're not going to pass the interview, no matter if it's done with AI, from memory or with stack overflow.
So what are we afraid of? That people are going to copy paste from AI outputs and we won't notice the difference with someone that really knows their stuff inside out? I don't think that's realistic.
Candidates could also have an AI listening to the questions and giving them answers. There are other ways that they could be in the process without copy/pasting blindly.
> To me it's always been about how someone reasons, how someone communicates, people understanding the foundations (data structure theory, how things scale, etc).
Exactly, that's why I feel like saying "AI is not allowed" makes it all more clear. As interviewers we want to see these abilities you have, and if candidates use an AI it's harder to know what's them and what's the AI. It's not that we don't think AI is an useful tool, it's that it reduces the amount of signal we get in an interview; and in any case there's the assumption than the better someone performs the better they could use AI.
Someone asking: "solve this problem" vs "what is the difference between array and dict" vs "what is the time complexity of a hashmap add operation", etc.
They give you different nuances on what the candidate knows and how it is approaching the understanding of the problem and its solution.
Also, I think that the people who are saying it doesn't matter if they use AI to write their job application might not realize that:
1. Sometimes, application questions actually do have a point.
2. Some people can read a lot into what you say, and how you say it.
Full quote here; seems like most of the comments here are leaving out the first part.
This feels similar to asking devs to only use vim during a coding challenge and please refrain from using VS Code or another full featured IDE.
If you know, and even encourage, your employees to use LLMs at work you should want to see how well candidates present themselves in that same situation.
If this is just for a written part of the process or something, maybe I get it? But even then, if you expect employees to us LLMs I'd really want to see how well they interview with the LLMs available.
As an anecdote from my time at uni, I can share that all our exams were either writing code with pen on paper for 3-4 hours, or a take-home exam that would make up 50% of the final grade. There was never any expectation that students would use pen and paper on their take-home exams. You were free to use your books and to search the web for help, but you were not allowed to copy any code you found without citing it. Also not allowed to collaborate with anyone.
Let he who is without sin...
Seriously?
Yes, certainly.
Over the years our skill at interviewing has grown. We try to craft a balance between asking questions which require exact responses, and those which the candidate is free to improvise upon.
AI may not yet be as good an engineer as most coders, but it's already absolutely much better at written communication than the average sotware engineer (or at least more willing to put effort into it).
But isn't this all the things AI promises to solve for you?
Isn't Anthropic's stated goal to make "nice" self-improving general AI? Cut out the middleman and have the AI train the next generation.
Maybe I read that wrong, but I suspect they are self-selecting themselves out of some pretty large talent pools, AI or not. But that application note is completely consistent with what they espouse as their core values.
Anthropic interview is nebulous. You get a coding interview. Fast paced, little time, 100% pass mark.
Then they chat to you for half an hour to gauge your ethics. Maybe I was too honest :)
I'm really bad at the "essay" subjects vsm the "hard" subjects so at that point I was dumped.
Oh, wow. I really believe they are missing out on great engineers due to the nature of it.
90 minutes to implement a series of increasingly difficult specs and pass all the unit tests.
There is zero consideration for quality of code. My email from the recruiter said (verbatim), “the CodeSignal screen is intended to test your ability to quickly and correctly write and refactor working code. You will not be evaluated on code quality.”
It was my first time ever taking a CodeSignal assessment and there was really no way to prepare for it ahead of time.
Apparently, I can apply again in 6 months.
For me the difference is that using an LLM requires a insane amount of work from the interviewer. Fair enough that you'd use Copilot day to day, but can you actually prompt it? Are you able to judge the quality of the output (or where you planning on just pawning that of to your code-reviewer). The spell checker is a good example, do you trust it blindly, or are you literate enough to spot when it makes mistakes?
The "being able to spot the mistakes" is what an interviewer wants to know. Can you actually reason about a problem and sadly many cannot.
Disclaimer: I work at Anthropic but these views are my own.
As someone who has recently applied to over 300 jobs, just to get form letter rejections, it's really hard to want to invest my time to hand-write an application that I know isn't even going to be read by a human.
If so, that's bad for their mission and marketing department, but it just puts them in the realm of a tobacco company, which can still be quite profitable so long as they don't offer health care insurance and free cigarettes to their employees :)
I see no conflict of interest in their reasoning. They're just trying to screen out people who trust their product, presumably because they've had more experience than most with such people. Who would be more likely to attract AI-augmented job applicants and trust their apparent augmented skill than an AI company? They would have far more experience with this than most, because they'd be ground zero for NOT rejecting the idea.
« The process we use to asses candidates relies on measuring the candidate's ability to solve trivia problems that can easily be solved by AI (or internet search or impersonation etc). Please refrain from using such tools until the industry come up with a better way to assess candidates. »
Actually, since the whole point of those many screening levels during hiring is to avoid the cost of having long, in depth discussions between many experts and each individual candidates, probably IA will be the solution that makes the selection process a bit less reliant on trivia quizz (a solution that will, no doubt, come with its own set of new issues).
So lacking any other ideas for a sample project I just started implementing Fizzbuzz. And IntelliJ started auto suggesting the implementation. That seems more problematic than helpful, so it was a good thing I didn’t end up needing it.
Write a program that describes the number of SS's in "Slow Mississippi bass". Then multiply the result by hex number A & 2.
Question 2:
Do you think your peers will haze you week 1 of the evaluation period? [Yes|No]
There are a million reasons to exclude people, and most HR people will filter anything odd or extraordinary.
https://www.youtube.com/watch?v=TRZAJY23xio&t=1765s
Hardly a new issue, =3
I don’t mind if applicants use it in our tech round but if they do I question them on the generated code and potential performance or design issues (if I spot any) - but not sure if it’s the best approach (I mostly hire SDETs so do a ‘easy’ dev round with a few easy/very easy leet code questions that don’t require prep)
If graphic design A can design on paper better that B, but B can design on the computer better than A, paper or computer, why would you hire A?
Anyway, bring back in-person interviews! That's the only way to work around this Pandora's Box they themselves opened.
However, not sure what to think of it. So AI should help people on their job and their interview process, but also not? When it matters? What if you're super good ML/AI, but very bad at doing applications? Would you still have a chance?
Or do you get filtered out?
The issue is that they are receiving excellent responses from everyone and can no longer discriminate against people who are not good at writing.
- AI companies ask candidates to not "eat their own dog food" - AI companies blames each other of "copying" their IP while they find it legit to use "humans" IP for training.
Is the application being reviewed with the help of an AI assistant though? If yes, AI mediation is still taking place.
That is likely enough said right there. Keep looking for a company that has it's head screwed on straight.
Of course it doesn’t…
When applying for a college math professor job, it's understandable that you'll use mathematica/matlab/whatever for most of actual work, but needing a calculator for simple mutliplication-table style calculations would be a red flag. Especially if there is lecturing involved.
this is about application.
(I need to reevaluate my work load and priorities.)
Low lifes
Exact opposite of our application process at my previous company. We said usage of ChatGPT was expected during the application and interview phase, since we heavily rely on it for work
If the interviewer points you at a whiteboard and asks you how to reverse an array, most likely they're checking you know what a for loop is and how to index into an array, and how to be careful of off-by-one errors. Even if your language has a built-in library function for doing this, they'd probably like you to not use it.
If the interviewer hands you a laptop with a realistic codebase on it and asks you to implement e-mail address validation, they're going for a more real-world test. Probably they'll be fine with you googling for an e-mail address validation regex, what they want to see is that you do things like add unit tests and whatnot.
You must have missed out on hiring some very good candidates, then.