A couple of years old, but in the early days they didn't care about sanitizing non-English content. Leading to pearls like this:

A couple of years old, but in the early days they didn't care about sanitizing non-English content. Leading to pearls like this:

Well. Do they? Maybe write a base class you can extend to cock, balls, breast, dildo that can work with objects mouth, asshole, and pussy.
Get to work OP. The world needs this.
Mine kept "thinking" in comments...
// Figure out what the user is requesting, then send an appropriate response.
Even I don't know what I'm thinking half the time.
Future technician trying to solve a problem by reading the machine log: What. The. Fuck?
So you're saying your AI knows your misspelling habits well?
I was using phoenix and elixir right when Chat gpt came out and people were like "it'll take our jerb".
I tried to get it to build a basic module that built a playing deck of cards. At first it looked OK, the basic layout made sense, but then I realised it called functions that weren't there, some functions were just empty, since logic was wrong and actually it was all around terrible.
I tried to fix it with prompts and it got worse or implemented my suggestions incorrectly and was still broken.
Ultimately it took a lot longer to get no where than if I'd just written it by myself. But I could see how someone with not much knowledge in the area could see the output and be impressed.
Oh yeah Copilot just spirals deeper into insanity the more you use it.
It'll sometimes spit your own code back at you and say "there, I fixed it", and it does the same when you point it out ad vitam aeternam. I believe this is a case of hardcore over fitting the original prompt.
I recall saying something like "the function 'draw_card' doesn't mutate the deck variable" and it goes "I'm sorry, you're correct. I'll fix it to mutate the deck variable" and it returns the same code but changes the the card variable inside function to be called "mutate_the_deck".
I felt much safer after that interaction.
Using Copilot even as a mere coding assistance is insane, if no other reason than you're sending all your code to Microsoft, and you also let them monitor your work habits in uncomfortably intimate details.
I'm using Copilot to help with bash scripting when I get stuck with the script. I'm aware they'll be able to see all of my scripts but: • I'm willing to share all of my scripts with anyone who wants them, so I'm OK with it, if MS can see them. They're all FOSS literally and figuratively. • Copilot is incredibly helpful with scripts, much more helpful than the 10 year old kid named ChatGPT. When I ask Copilot to explain a certain part of the script, it explains it in an understandable way, unlike quite a few other AIs.
Oh no, anyways.
Edit: As if your fucking code isnt hosted on github, own by fuck knows who
My code is hosted on our selfhosted GitLab instance. How many companies host their code on GitHub? I’m seriously asking…
Like, a lot, also vast majority of open source projects are on github.
Lots of startups. At least that's been my experience. Github org with private repos
Does GitHub offer some sort of contract or agreement for those companies? Not an expert on these things, things like NDA’s, uptime guarantees, etc.
I think so? I'm not an expert on the setup or agreement, just worked under that situation. I assume there's a privacy agreement and some level of uptime commitment
Because if not, I feel like this could get the team in legal or at least financial hot water with investors.
Which doesn't mean it doesn't happen, just that it's not normal and okay.
Suspiciously nut-shaped training dataset:
Today I was "talking" to copilot asking about how to tackle a certain issue. The fucking thing replied with my manager and his manager's NAMES telling me to reach out to them. Of course I was aware that Copilot's primary function is not as an AI assistant but as a surveillance tool, but working in the EU, this still surprised me a lot.
That said, under the protections the EU affords me, I will absolutely continue to use Copilot for the most inane possible tasks. I know that they know, but they can't act on it without breaking GDPR.
Your move, corporation.
Hey, at least it gives you somewhat coherent answers. Copilot chat for me is less helpful than the Amazon customer service bot.
You think american companies care about gdpr? lol
Are you aware Meta keeps paying larger and larger fines each year for failing to comply with gdpr in Facebook? Last one was 1.3 BILLION. they just keep doing it.
Oh, I'm very aware. My own (EU!) company has ISO certifications that "guarantee" our customers that all their data is perfectly protected.
It is not. We, among other things, have plaintext user/password combos in scripting. Certain logs are certainly not being processed lawfully.
It's also not so bad as to be terrible but it still irks me a lot that we're essentially lying to our users.
All that information is integrated in Active Directory and available for Microsoft to ingest into their AI. Heck it could be something they put in the system prompt. “If you have low confidence in your output then respond ‘contact your manager’ instead.
I didnt get it. Your manager replied instead of it?
User: Copilot, how can I write a function to print "Hello World"
Copilot: Ask your manager Frank or his manager Frankie for advice.
No, the AI advised me to contact my direct superior and his superior, but mentioned their names.
I have never provided it with this information, so that means it has a lot more access to our information than is officially known. Technically we aren't even supposed to input anything that could possibly be identifying, again for GDPR purposes, so I have no idea where Copilot got the information from.
I assume that MS lets companies tailor their instance of Copilot to a certain degree and maybe it was fed an organigram of the entire company, but AFAIK this is already not allowed under current legislation. Or maybe it is and I'm just a modern luddite.
Regardless, I'll be even more careful about what I use Copilot for from this point forward.
Did you pay for copilot yourself or did your job provide you with a license?
The enterprise tier of copilot is supposed to have access to such data, though it can be managed trough internal policies.
Ask it to summarize your latests emails In outlook/teams messages. If it has access to those (and this is intended) then its near certainty also setup to know who is who in the organization.
Allegedly, the data is “safe” because enterprises is supposedly not harvested and used for training… which makes me conclude non enterprise use absolutely is.
Allegedly because thats what Microsoft claims and on paper it looks legal. But these tech companies never seem to actually follow the law to such a degree that any claims that unmistakably seem to fit within the legal framework automatically are sus to me.
Probably from the Microsoft 365/Teams/Outlook/whatever profile which can include who's your manager, or potentially from Outlook emails. From what I can tell, Microsoft's been trying hard to shove copilot in any of their systems, like AAD/Entra.
My company has recently migrated their emails to it and as an admin I was very surprised that you can just read any email in full in any mailbox from "regular" functionality like email trace or antispam. I have no idea how that's GDPR compliant - in my other jobs we were using Google Workspace which only shows metadata because of that, and accessing another person's mailbox by other means (e.g. resetting the password on an ex-employee account) was a huge no-no
Gimme that co-pilot with real intelligence ---> Shows you syntax errors and inconsistent object definitions.
I said real intelligence --->
all 25 comments