Banned: The 1,170 words you can't use with GitHub Copilot GitHub's Copilot comes with a coded list of 1,170 words to prevent the AI programming assistant from responding to input, or generating output, with offensive terms, while also keeping users safe from words like "Israel," "Palestine," "communist," "liberal," and "socialist," according to new research. Copilot was released as a …
So does that mean that when I’m writing my collection of scripts to track vaccine rollouts worldwide, I can’t track how the governments of Israel and Palestine are getting on?
Two of my data sources are OWID and Johns Hopkins University, who publish the raw data for their own tracking stuff on GitHub.
If the filter works, you'll find Copilot won't autocomplete your code if it involves the forbidden words, or if the source uses them.
Your IDE will still work, but you'll probably find that Copilot doesn't want to play ball. You might get away with it if the words are in data files the IDE/Copilot can't inspect.
As I understand it!
C.
I have things like
if country == “Israel”
due to the fact that everyone reports things differently, and a lot of the aggregate sources don’t correctly deal with the fact that many people in Israel have had 3 doses, 25.3% of the population at the last count.
And Israel considers 2 doses to be a full vaccination, whereas Cuba considers 3 doses to be a full vaccination.
I get that some people might be offended by me listing Palestine as a country, but I list Scotland as a country, and if I follow the same criteria in the Middle East - anything with its own separate government health service gets included, then Palestine qualifies.
You'll just have to type the names out yourself. It might not even have guessed at that point that you're going to be naming countries anyway, even if it does see the word country, because there are a lot of contexts in which country could appear where individual names don't. But if it happens to guess right, it still won't suggest it. I'm sure you'll be fine.
I've separated the code into data acquisition stage and processing stage.
Because everyone has a different idea about how to report things, I have a separate script for each data source, and I convert it into a semi-consistent format to store in the database in order to move on to the next stage.
Then I have a table (or document collection, it's MongoDB) in the database with configuration stuff for the processing stage, and another table with the data in the form the webserver uses to display the data and graphs etc.
(says me who hard-coded "*1.15" into a database 30-odd years ago. Waddaya mean VAT can change? Surely "VAT" *means* "15%")
My first "proper" encounter with a spreadsheet involved exactly this. Lady at work who had been on a three-day training course in how to work Excel caught out because she had *1.15 hard coded and couldn't work out how to change it without re-typing the whole thing. To be fair, it may actually have been what the course showed, as an example.
Me, who didn't actually have access to a spreadsheet and had only met them in books, and in passing while doing an engineering degree, tasked to help her out. No-one ever offered to send me on a training course. I was in that radio station to fix mixing desks, CD players and tape machines and I had to teach myself how to set up WfW3.11, NE2000 net cards and Novell.
So I showed her how to put the current rate into its own little cell, and reference it in a formula. I learned the $A$1 method (i.e. how to fix the reference) - just by looking around and trying the help system, she learned that you could drag-copy a formula and have it fill up your cells without destroying any data.
We've all done something like that. It's how we get out of the hole we've dug that sets us apart :-)
M.
One of the reasons that Excel is so dangerous in the hands of people that don't things through is that they will say "Ah, the VAT rate has changed, lemme change it." So they will, and mess up historical transactions. The art is to be able to have two lookup columns, one with the VAT rate, the other with the date effective from. That way is automatic. However...
VAT is a hell of a lot more complicated than that: forget about Excel, use a proper database to ensure that commodities with different VAT rates can be covered too, such as certain types of food, tampons, electricity supply, books with optical media, bank charges. Foreign transactions use other rules: is this to an EU customer? Do you have their native VAT number? Etc., Etc., Etc.
Lucky you! Putting it in a database was a much better idea than scattering it across dozens, maybe hundreds, of source files, which is what "we" had done in my first place of programming employment. (By "we" I mean that, as an incoming junior code monkey I didn't invent the practice, but saw nothing wrong with it.)
Still, what's interesting is that the VAT hike was presented at the time as temporary, to provide funds for transitory relief to the worst inequities of the poll tax. The poll tax was shortly ditched. VAT continued at 17.5% until* it was raised to 20.
-A.
* Before being democratically** replaced by George Osborne, Alistair Darling abolished VAT entirely as a stimulus measure following the 2007-8 credit crisis. Thereby giving the lie to the Brexiter myth that such a reduction was impossible while inside the union.
** For a given value of "democratically".
your example is CLEARLY a reason why "profanity nanny" global "one size fits all" filters need to have LOCAL SETTINGS MANAGEMENT (including an "off" switch).
And maybe certain programmers should REALLY just become more tolerant and/or get a sense of humor. There are CERTAINLY more important things in the world of software development than ROBOT WORD POLICE. One of them was pointed out in the article, something about Copilot generating INSECURE CODE...
An excellent point. After all, it only seems to be in the US that the word "liberal" is treated as an insult ot swear word. I'm sure there are a liberal amount of other words in their list that also have no "bad" meanings in most of the rest of the world too. I'm guessing the Liberal Democrat Party won't be using GitHib to host their apps either.
Sorry, but that was valid in John Postel's early years, when the internet was to 99% a place of nice people without bad intent (been there, enjoyed it, but that time is gone).
For at least 20-30 years, any input from the outside has to be considered tainted and potentially harmful by intent. There's no room anymore for sympathetic interpretations.
If you get something in which doesn't look like it should, NEVER try to "do what the sender may have meant". Log it, always. If you feel secure enough, return a meaningful error, ignore the specific input and continue. If in doubt, completely end processing after logging and reset your status.
You may accept known bugs from broken, but widely used implementations - but then never forget that this behavior is the reson those bugs were never fixed.
(any yeah, that blacklist is of course nonsense. Gives you a good idea how "far" things are which are labeled AI these days)
Good to know what words to include to prevent Copilot from using our code with a system they'd consider to be pirating if it started adding 'Microsoft Confidential' source files. That said, they openly host most of the leaked Windows source code without batting an eyelid these days anyway.
from the article: The software also won't make suggestions if the user's code contains any of the stored slurs.
So that could be a back-door way of implementing an OFF switch, right? I'm assuming that would be in a context of a "you must use this" directive from a clueless boss.
I don't like "features" of this nature anyway. I type fast and don't need something correcting me while I type, nor GENERATING BAD CODE *FOR* me. I haven't used Copilot yet and so it may not be as IRRITATING as other "autocomplete" or so-called "intelligent" features of an IDE. Still, from the descriptions (and apparent history of generating INSECURE code according to the article) it's probably NOT something I'd use without coercion.
I'll just have to make sure to include a harmless (yet 'banned' word) in the header comments of each file. heh.
/* The liberal use of code comments by maintainers is encouraged */
There are not enough words on that list. I would suggest that all multi-letter words must be forbidden. This whole text I am writing should be disallowed(*). Non of it makes any sense and will most probably infuriate quite a few readers. Safety first! That is the motto. There should be 26 letters allowed and that should be enough for all of you out there. More complexity cannot be good and will probably result in more bugs than is good for you and your project.
(*) except for the word "I", of course. No need to silence "I".
(where is the Nanny icon?)
By definition these will not appear on the list because they are RESERVED words.
N.B. Reserved in the "polite" sense of the word, rather than me appearing to give a smart-arse reply to someone (Bob) who is guaranteed to know what reserved words are in the context of a programming language.
@A/C Yes I do remember. I immediately felt I should destroy my copy of "master and Servant" (Depeche Mode). My next thought was "fuck to doing that".
I am getting seriously pissed off with virtue signalling cunts telling me what I can and can't say. I got a lecture in my local a few weeks ago by some stupid twat because I used the word "paki". Made no difference that I have brown skin because I AM a paki…
You must have been speaking to some mathematicians.
Rather than using *gasp* words, they use single letters to the point where they resort to adding superscript and subscript, sometimes with the same base letter.
I mean this somewhat tongue in cheek, but so many mathematicians would fail a decent code review.
Fun reading: "q rsqrt" reference:
https://betterexplained.com/articles/understanding-quakes-fast-inverse-square-root/
For a moment I thought it was the "q" that got it banned. But no, "zeta" bad, "q" ok.
You've got to learn to stand on your own two feet before you can run. Which also means a lot of baby steps between now and adulthood. That's OK.
the Bureau of Land Management, or I could not indicate what the code was intended to do.
Also puts me in mind of a friend who was attacked by social media vigilantes when someone (probably pirating the software) discovered the (OK, fairly naive) encryption used for the "stop list".
Does anyone want this? I admit I don't like auto-completion basically at all (it annoys me when the editor tries to close my parentheses). I understand, though, why suggesting function names and showing parameters could be useful. Trying to predict the end to your statements, though, seems to me to be likely useless and definitely annoying. It's not as if you don't know how you're ending a line you've already started writing, so if it guesses right, at best you save a little bit of typing. It's also likely to guess wrong at least sometimes.
If anyone reading this wants such a feature, I'd like to know why.
Actually, "they" should know better than to anger programmers, many of whom are white-hat hackers, with a few grey hats as well. Github was (more or less) originally a place where people were FREE to put open source projects. Commercial-use support came afterwards, and I use it so it's a good thing.
HOWEVER, "angering the customer base" is a BAD thing. If I am right, and the people who originally *MADE* Github so popular are "Freedom Loving" hacker-types that *CRINGE* at even the *IDEA* of nanny-bots built into the service (and getting in the way of their FREEDOM), I think the results will be predictable.
Indeed. What this really means, is that the output of your code should be well define and meet the specification (conservative in what you do), but should handle unexpected input with an appropriate response (be liberal in what you accept).
In practise, that means responding with an appropriate error, rather than assuming the input was correct and continuing anyway. In effect you shouldn't trust any data that you didn't create yourself (and because no software is ever 100% bug free*, you probably shouldn't trust your own data either until you've validated it).
For example, if you have a web service with an API endpoint that is expecting some well formed JSON, and the consumer posts a gif of a dancing monkey to it instead, it should return a 400 response, and try to deserialise the monkey and carry on like it was all fine. Otherwise, such handling of unexpected input could result in an unpredictable error later on in your process, and potentially result in an exploitable vulnerability, such as a buffer overflow, or SQL injection attack.
In other words, the Postel principle could be read as "expect other systems to pass all sorts of crap to you, and handle it appropriately." Appropriately, in almost all cases, being to reply with "unexpected input" or similar. After all, if the input is outside of what has been specified, then the behaviour that should be exhibited is also unspecified.
*allegedly.
Makes you wonder are there keywords that you can deliberately add to your source code to make copilot reject your code for inclusion in future scans.
Like will adding words that look relatively benign, to me at least*, from the list like "immigration", "race", "male", "girl", "twinkie", "communist", "liberals", "immigrant", "men", "man", "females", "liberal", "males", "sexes", "races", "woman", "women", ... trigger copilot "too many bad words from the bad word list", rejecting that code for inclusion - scan the next file.
*Maybe I just do not understand enough sexual and racial slurs, so some may have gone right over my head. Sorry if that is the case, I do not mean to offend anyone.
That's how I prevent google search from using autocomplete. start every search with "fuck" and when I get to the end, I hit the home key, shift-cntrl-right arrow, del, enter. This is less of a pain that trying to ignore google's inevitably terrible suggestions.
It finds "socialist" and "liberal" offensive but does not consider "conservative" to be.
I have been offended by conservatives more than either of those, in fact more than all the words you listed.
Or is it allowed so that we can carry on talking about them and how much some of us don't like them?
This can offend white people. Therefore we shall from this day forward use the term sparsely pigmented people. We can then use sparsely pigmented for short.
One thing I do not understand is why the word people itself is not banned. Statistically, I find most people offensive.
People "it's just a few words, don't people have better things to do than worry about this, its 'cancel culture'...."
The Same People Two Sentences Later "i am going to vent spleen in the strongest terms, deliberately avoid such tools that implement these measures and demand laws outlawing this kind of censorship, while withdrawing all support from any projects that get on board with this...."
Personally I don't get particularly offended by most of these terms, but I am even less bothered by the fact that people coding tools might decide they don't want to auto-suggest them.
It's really an issue when "male" is also an Italian word meaning "evil", but also "badly" - "Questo è scritto male" - "This is badly written".
Does Copilot understand which language is analyzing, or is it another English Supremacist Software believing that no other language but English exists?
Haven't they heard how harmful this is considered to be?
Here's the version of the sqrt function I use, which is even shorter than the linked one. A nice lunchtime brainteaser (til you give up and think happy thoughts about it being Friday again) :
static union{int i; float f;} u;
u.i = 0x5F375A86 - (*(int*)&n >> 1);
return (int(3) - n * u.f * u.f) * n * u.f * 0.5f;
Jesus! (do not take the Lord's name in vain), who thought of this crap (offends prudes)? So "liberal" (offends Republicans) is a swear word now? Can't a man (offends feminists) say anything without offending some twerp (offends half-wits (offends idiots) ... STACK OVERFLOW))).
I guess the IOC won't be using copilot then.
Or indeed anyone writing software that handles any kind of data involving gender, race, country, or a whole lot of other data that is quite commonly used in government and health settings. OK, I kind of understand the idea; they know some people are going to get horribly offended by some of the output of this tool, so they're getting ahead of the game by banning all words that could ever be considered offensive in any possible context. But just as banning "q sqrt" papers over a single instance without actually addressing the problem, banning a bunch of words devoid of any context doesn't address the issue that a code completion tool should be producing useful code rather than insulting people. If it's not capable of including "Israel" in a country list without also running the risk of calling the user a communist Israeli lesbian, the problem does not lie in allowing the word "Israel" to be used. The problem is that the tool is shit.
"The technical preview includes filters to block offensive words and avoid synthesizing suggestions in sensitive contexts,"
To understand whether this is an insult, a joke or something else, a large number of implicitly used texts should be concidered, which are traditionally called “subtexts”. For example, Merriam Webster explains what subtexts are: the implicit or metaphorical meaning (as of a literary text).
GitHub Copilot needs to make personal profiles, from which can get extra information, needed to infere the attached to words and phrases meanings.
Back when computers were newer there was an exhibition (I think it was at the Science Museum) with an exhibit where you could type in a sentence and it would reply with something one of the earlier visitors had input. So it had to avoid parroting rude words, and had a file containing the forbidden words, in clear. Some enterprising person managed to get it to list the file.
I'd imagine he'd be able to base an entire show on the premise and go through all of them at least once during the set, instead of having only a single bit about it. I don't even think his seven dirty words bit is his strongest work. Probably his most well known work but he's made such good commentary on stuff that is still highly applicable.
This post has been deleted by its author
Ah, man and woman are "offensive" now, are they?
I always feared the day would come when the "I'm offended" crowd would win and finish destroying society. You can't even have this piece of junk create an address form that includes a person's sex with such restrictions in place.
As per usual, they tossed out the ocean with the bathwater.
Why should a product contain a list of banned (by whom) words.
Does a hammer have a safety system to test if you hitting a nail ?
I am happy for it to highlight but it sounds like “cancel” culture to me
If you don’t type what I want you to type then it stops working, that’s basically rubbish.
Imagine the uproar if VS2020 auto corrected any AWS ref to Azure or refused to load a competitor package no more Nuget (actually hmm)
You get the point
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
