RE: https://mastodon.social/@mcc/117553485333745988
If you’re in Massachusetts, 👇🏻trust me, you do NOT want the top-two primary; we have it in Washington and Two Rs vs Five Ds is the only way Rs get voted to state offices anymore
I feel like maybe having a private company control such a high percentage of tech for public schools was a really really bad idea (see: Google search, Google classroom, Google Docs/Sheets/etc, Google Chromebooks, Google Drive, etc etc)
Time to ditch Google and embrace other solutions!!
What you'll find though is that the vibecoders / slop artists will be a bit bitter that they can't participate in the "main" event. They want the respect of the place that excludes them, because of course, Claude vs Claude *isn't very interesting*. It's much more fun if there are a bunch of actual artists around, so you can elbow them out, so you can look good in comparison to them.
Microsoft just gave us the perfect name for AI thieves: Kyle
I have been thinking a lot about this 30 second advertisement release by Microsoft, because more than anything I’ve seen it summarises what generative AI is actually for.
First, please go watch the full thirty seconds if you haven’t already:
This ad is produced by Partizan (their terms and conditions prevent AI copying of their work, by the way) and Division Film. As I said on Bluesky, it’s an amazing piece of work. I hope it gets archived somewhere forever.
It’s an open admission from Microsoft about the purpose generative AI is designed to serve. A mediocre white guy sees his non-white colleague getting praised for good work, he asks Copilot to literally copy it, and then he gets all the praise instead. There have been so many ads for generative AI that have been confused, misleading and vague. This one just tells a specific product truth like no other piece of advertising ever has before.
Let’s do a shot-by-shot breakdown of this absolutely phenomenal piece of cinema (which, by the way, was shot on film).
Shot-by-shot breakdown
00:00 -> 00:05
“If you ask Kyle, he wouldn’t say he’s competitive……………out loud”
A little bit of scene setting. Kyle’s an unassuming office worker, in a boring office. He’s got a warm smile, and gives a friendly nod. “Hey!”
The narrator, milliseconds into the clip, reveals Kyle’s frazzled mental state in two words: “…….out loud”. Millimetres below the surface of his regrettable facial hair, a roiling ocean of suppressed rage.
00:05 -> 00:11
“But at 9.42am, Daniel’s forecast model was called……’Honestly, Beautiful'”
The narrator drops a precise timestamp, like he’s recounting the exact moment of a terrorist bombing inside Kyle’s soul. We see Kyle’s colleague praising Daniel’s forecast model in the chat, and there’s a shot of colleagues crowding around Daniel’s desk, clearly awed by his forecast model.
And we see the coworker’s words reflected in Kyle’s eyeball – a popular horror movie trope. Do not doubt that the first moments of this ad are a horror movie. This is Kyle’s nightmare: a Black colleague receiving praise for good work; surrounded by adoring colleagues. Unjust and unfair; Kyle deserves the applause. Kyle could never explain why, but he knows that he was born with the right to be the main character, and Daniel’s accolades are an unacceptable and unnatural transgression.
The racial politics of this ad are fucking incredible. And my god, the number of people in real life I’ve worked with who have this exact horror movie playing out in their heads whenever they saw some foreign pretender like me enjoying success. Let’s just say this ad is clearly based on a many true stories. Absolutely the same applies to gender politics too, going by the replies I had when I shared this video.
Not just a horror movie trope by the way, but the opening shot of Blade Runner; a film in which human-created biological beings show more verve, more life and more vividness in their eyes than their dull creators. Kyle’s pain calls back to the vivid, glowing inner life of the Replicant, and the jealous, sweaty inadequacy of human beings.
00:11 -> 00:16
“And honestly, that got to Kyle”
This is a brief sequence but it’s the most important in the whole ad, because it really depicts what ‘got to’ means. A rapid sequence of shots, each lasting only a few frames, shows
Kyle’s pinballing mental state, careening between joy, tear-sodden sorrow and blank-faced neutrality, has a cure at the end of this sequence – the Copilot button in Microsoft Excel.
00:17 -> 00:21
“But no worries, he just turned to Copilot”
Kyle’s real smile comes roaring back, when he types (as implied by the text overlay) into Copilot:
“Analyse Daniel’s model in the all hands channel”
“Improve my charts”
“Make mine beautiful”
The last of these is overlaid on top a of a tight close-up of Kyle’s grinning lips, whispering the words out loud:
My charts. Make mine beautiful. It was Daniel’s model, but it is now Kyle’s, turned into something “beautiful” with a whispered command. This, my friends, is the fundamental core purpose of generative AI. It is the tool of the mediocre, jealous colleague, who uses your work as input into the machine, to spit out a garbled version and slap their name over it. It is aggressive, non-consensual assertion of ownership over someone else’s work.
It is not just stealing someone’s handbag: it is walking away very slowly after you’ve stolen it, yelling “hey, I MADE a HANDBAG”.
And the Kyles that steal like this do not recognise it as anyone else’s work: they genuinely, sincerely believe it is theirs. I see this all the time on social media, most of all LinkedIn, where someone announces they’re sharing the outputs of generative AI, and describes as “Hey everyone, I made this….”. Claude, Copiliot and ChatGPT are programs tuned to twang the strings of male ego like no other piece of software in human history; puffing up users with dripping compliments. Of course they feel like they did it themselves.
00:22 -> 00:25
“You see, Kyle’s not competitive. It’s not even a contest”
Out comes the dashboard (looking exactly the same as every other corporate dashboard has ever looked, by the way. It’s not beautiful!). Daniel is looking at Kyle like this is the twentieth time that this has happened this month:
kyle you piece of shitAnd that line, holy shit. “It’s not even a contest”. Of course it isn’t. Again, this is the core function of generative AI.
Daniel’s role in the right way of things should be that he is the feedstock for the machine that spits out Kyle’s ‘work’. Daniel is an input, not competition or a colleague or even really human. If it were a contest, Kyle would lose. The only way for Kyle to win is if there’s no fight at all. Daniel always loses, Kyle always wins.
00:26 – 00:30
In the final shot, a swarm of colleagues fawns over Kyle. They chatter excitedly above and around his head, but he barely hears a word, he’s so lost in his own success. A woman rubs his clothes with a lint roller; an obscenely gendered show of submission and reverence. Another colleague simply sits at the bottom-right corner of his desk, head propped up with his arm, gazing lovingly and unbreakingly at Kyle.
Kyle leans back and for the first time, displays a genuine smile. Nothing is simmering below the surface. Kyle is at peace. The world is fixed.
This is what GenAI was always meant to do
I saw a few responses that theorised the agency was doing ‘malicious compliance’ here; intentionally making an embarrassingly honest ad. I don’t think that’s true – I think this was meant to be a short skit with humorous overstatement that still showcases a sincerely-held application of the product being advertised. There are jokes, but they still meant it. It would’ve also passed through a shocking number of layers of review before being aired around the world (many people who responded said they saw it in the UK, for instance).
As I have detailed on Bluesky, something really notable is how AI plagiarism is not just our life’s work being fed into models to use as training data. Despite the many megatonnes of fossil fuels poured into the maw of chip-meltingly hot data centres, the raw outputs of modern models are still just…….fundamentally fairly shitty. Kyle has to beg Copilot to make his thievery look ‘beautiful’.
It makes way more sense for content thieves to feed new work directly into a model, and request a perturbed version that’s different enough to evade copyright laws or easy detection. Most recently, comedians (usually young women) have been having their work stolen by people using generative AI to create semi-pornified versions of the stand-up sets.
In one recent case of thievery, Fred Lambert of the rapidly slopifying cleantech site ‘Electrek’ simply fed one entire blogpost written by me into an LLM, and produced essentially the same post but with zero writing quality, data visualisation skills or connection to real, material issues. Ironically, of course, it was a post critical of Elon Musk’s AI data centres. Electrek probably earned a decent amount of revenue from my work by running ads in their article. The spectre of a mediocre white guy using AI to steal my work to pretend to be against AI is just…..a lot.
https://bsky.app/profile/ketanjoshi.co/post/3mg6jy2e6kc2u
There is essentially nothing I can do about this sort of thing, besides suggest anyone who hasn’t already completely blocked Electrek to make sure they do so. It’s the same story no matter where you look: there is absolutely zero recourse for the victims of generative AI assisted theft. Social media platforms could easily choose to aggressively crack down on it, and bosses in corporations could choose to treat it as a punishable offence. But in both instances, they very much benefit from the newfound permission space for theft created by AI hype, so the chances either will take action is close to zero.
He is a Kyle; he just did a Kyle; I think this content is Kyle
With all of this in mind, I am inventing a new word. Kyle.
Kyle can be a proper noun; a taunt. There’s a good chance every real-life Kyle looks pretty much the same as Microsoft’s fictional Kyle.
Kyle can be a verb, too. Oh my god, I think he Kyle’d my report. And finally, we can use Kyle to describe the finished ‘product’. Man, this video looks really Kyle.
Kyle is the entire lifecycle of AI-assisted theft enacted by mediocre white guys who’ve always stolen the work of young people, people of colour and women their whole lives, but who now have access to a machine that does it breezily and effectively. Kyle is the person, the act and the output. Kyle is the feeling: the deep, unchecked joy he feels when he’s cheated his way into success rather than earning it. Kyle is the wind in his hair he gets to feel because he’s standing high on a bunch of other bodies.
If they’re saying it out loud now, why don’t we say it back to them, and reintroduce some the normal moral punishments we use to dole out for thieves and free-riders? Your boss isn’t going to do it. The social media platform isn’t going to do it. So maybe we’ve just go to do it ourselves. Watch out Kyle, Daniel’s pissed off.
hey, a level is coming together! #lispgamejam
for the love of everything, if your interface for a numeric value has a slider and a numberbox but the slider provides greater precision than the numberbox because you’re rounding, just, don’t ever do any interface design ever again please #lispgamejam
my first time making tiles for a game outside of tic80 and if you haven’t done this before I do NOT recommend isometric for your first go around
I’m happy with what I’m doing but slicing tiles from my graphics program while simultaneously lining everything up just so is a real pain
Billionaire says it’s “quite terrible” for the public to have money; private equity offers much better utilization
What are YOU looking at this #caturday
Whenever the tech news says "new macbooks omg so much faster", my brain hears "web about to get slower again."
first rule of asyncronous ADHD interface design:
there is no such thing as async ADHD interface design. if something doesn’t happen immediately, it’s either not happening, or it needs to do something else when it’s ready that can be responded to
this message brought to you by an in-web-page support chat that I simply forgot about for several minutes while the person had to look something up
I no longer have even a single chat app where it is possible for me to understand "when you beep at me, how do I tell why you beeped at me?" or "what do I do to make the 'unread notifications' tag go away?"
Helping the kiddo get their first pair of glasses and having to explain gender coding in frame stylings, something I had never actually *thought* about in the forty-plus years I’ve been wearing glasses
Periodic reminder: EVERYONE. Even very smart people. Even people who work in security. Even people who work with account phishing. Even you. EVERYONE can fall for a scam.
The more strongly you believe you could never fall for a scam, the worse you're going to get fucked when one slips past your defenses. Scammers LOVE people who think they're too smart to be caught.
Never shame someone for getting caught by one - be understanding and helpful and supportive. Because you're going to be in that seat someday.
We are in a very dangerous time, when "AI" security companies are competing to prove that their technology will provide returns on investment. The industry has always been broken. There has always been an incentive to inflate the importance of vulnerabilities, but there was always a risk of burning hard earned trust. In the end, a lot of folks decided it wasn't worth it.
But "AI" companies haven't worked to earn that trust. They have something to prove, right now, or they don't exist. So now everything is critical, "impossible to find without AI," and so on. Blah2shell sells, and now everyone wants their little vuln labeled that way regardless of the complexity or risk.
Burning trust in security is the absolute worst thing you can possibly do in terms of tech. I'm worried about where things are going right now.
There have always been more bugs than we can fix. We've had to triage. If we break that, we flood maintainers until they burn out. Our job is not to fix everything. It's to filter information for the people who do.
But right now I'm reading vuln reports that aren't distinguishing between a 100% reliable and trivial exploit like log4shell and complex memory corruption bugs that are incredibly noisy and basically impossible to exploit with ASLR. ASLR is like 25 years old. You have to do better than a simple buffer overflow now to get a shell.
But people have shit to sell, so we have more noise. Again, "we've put the attenuator on the wrong side of the equation."
more than any of that future-looking stuff though it fucking sucks right now at this exact moment. it is actively shredding trust between teachers and students right now. the current class entering college was mostly in 6th grade when covid started. it’s a cohort that has undergone so much during such a formative time, they need and deserve so much care and support, and we’re giving them fucking chatbots. it’s such a failure. it makes me so sad
open source, lately
normalize the idea that general-purpose software doesn’t require a fucking background process
normalize the idea that general-purpose software doesn’t require network access to do things that shouldn’t require the network
normalize the idea that general-purpose software doesn’t require administrator access to run on your fucking computer