▲ 70 points
back
48 comments
To be technical, what he read wasn't prompt, it's part of the AI's response to the prompt.
But yes, it's a giant tell that politician used AI to write his speech, and he didn't thoroughly edit the response when preparing the final draft.
No, that politician trusted some poorly paid staff to write a boring procedural speech while he was doing something (hopefully more useful).
The staff used an LLM and did not bother to check (or got their check wrong.)
One of them is guilty, the other one is responsible, the conséquence should be in proportion to the gravity.
There's a lot to unpack here.
> that politician trusted some poorly paid staff
- I assume that politician is the one / main person in charge of doling out payments. Does that mean they share guilt/responsibility for the result?
> to write a boring procedural speech while he was doing something (hopefully more useful).
- Is procedure not his job as a politician?
- If not, what is his job / the more useful thing?
> One of them is guilty, the other one is responsible, the conséquence [sic] should be in proportion to the gravity.
- Which is which?
- Which has more gravity and why?
- What consequences are you envisioning, in both cases?
- If, as I suspect, the consequences are lesser (proportionally) for the politician himself, what if any deterrents can be put into place to stop that behavior? After all, he manages his staff, in theory.
Of course they share responsibility. Pay scraps for slop, expect slop.
The more useful job of a politician may be similar to how managers in general have a "more useful" part of their jobs then publishing status report.
"Gathering information, making decisions, nudging others." Lobbying for their constituency (at best) or the people who paid for his campaign (at worse).
Or, heavens forbid, writing legislation on a topic that he knows something about, as opposed to whatever crap he had to slop about.
As I said, responsibility has to be shared. I would want neither the staff to be considered the sole problems, nor the polician career to be permanently halted because of this. (I know he's going to look stupid on the internet, and we'll all have a good laugh, but maybe he's also working on some serious stuff, and real people are going to pay real consequences because of what is, in the grand scheme of things, a minor blunder.
I'm more concerned about students not caring to learn to read because of AI.
> [post 1] "One of them is guilty, the other one is responsible"
> [post 2] "Of course they share responsibility."
I'm not following.
We're also still disregarding the fact that none of us can even prove that it was one of his staff that did this and not him, but your OP is worded like it's a given it was staff.
> The more useful job of a politician may be similar to how managers in general have a "more useful" part of their jobs then publishing status report. "Gathering information, making decisions, nudging others." Lobbying for their constituency (at best) or the people who paid for his campaign (at worse).
> Or, heavens forbid, writing legislation on a topic that he knows something about, as opposed to whatever crap he had to slop about.
Not sure about you, but I often find my job will oftentimes not always include the things I have the most experience in, and I'm expected to be able to rise to the occasion because I was hired to do so.
> As I said, responsibility has to be shared.
You said that? Because I read "one of them is guilty, the other one is responsible." If you mean that responsibility is synonymous here, why the distinction?
--
Revisiting my original questions that have gone unanswered:
- Is procedure not his job as a politician?
Given that you appear to have instead answered "If not, what is his job / the more useful thing?", I'm assuming "No", to which I disagree. If your answer is that "yes, procedure is part of his job", then I assert he failed to do his job here, regardless of how boring you find that job to be personally.
If the guilty/responsible distinction still holds, I'm still unclear as to what applies to whom.
The slop jockey failed to deliver a cleaned up speech. The politician failed to proofread it or even just skim it once before giving the speech. The journalist failed to understand the difference between a prompt and LLM output.
Every single level failed here. Nothing works anymore. No one cares about anything making sense. I want fucking out.
> "bro there's no way he's gonna read this"
< "cmon do it bro it'll be so funny"
"Don't attribute to malevolence what can be adequately explained by large amounts of stupidity."
Hanlon's Razor
Encouraging that news.com.au (a Rupert Murdoch shop) staff can't tell between AI prompt and response...
https://www.youtube.com/watch?v=3MU6q9G0Sf0
I would rather speak to this man than anyone reading anything AI-written
For those not grasping the joke: the response is a largely incomprehensible mix of Hindi (I think) and (very-heavily accented) English. From an Indian news programme.
Shades of "Prisencolinensinainciusol":
Many comments pointing out this is part of the response rather than the prompt.
But the whole speech is the response. What do you call the bits at the start "Certainly, I can write a speech in this style" type thing, and at the end "Would you like to learn more about political speech writing?"
Perhaps something like "AI tells" but perhaps there's already an industry term.
Edit: hadn't watched the video, this is the former "response to prompt" type tell.
> Here's a more natural flowing version of that section that reads like a legislative speech rather than a series of short points.
That is the video title, but I think it should be paraphrased here to be more correct given our highly technical audience. The politician is actually reading an unedited AI response, which includes "editors notes" or maybe you could call it meta-content.
I have both experienced and hear my coworkers across the tech industry complaining about having to parse slop from their peers, management and juniors they work with. AI comments on their merge requests that sound right, but are subtly incorrect and irrelevant that add to their cognitive load. PRD (Product Requirements Document) slop from PM's that are adding features for the sake of it and is essentially busy work. Leadership asking to add AI to evaluate employees and dashboards out the wazoo.
The floor has been raised for sure, and I am writing less boilerplate than before, and am able to diagnose bugs in my work quicker, but the amount of people that seemingly just copy and paste the AI output without critically thinking is causing me distress. There are seemingly no consequences either, more AI = more good.
Policy influence idea: find out who are the assistants of policians and what ai assistant, subscription they use and give them different answers than to everyone else. You do not have to even corrupt them.
Now identify voting blocks amongst your users and do it on a national scale.
If he only reads the prompt and the RNG state of the prompt time it might be an improvement. Now people will claim that the RNG is sentient.
This is our new normal. It's not necessarily interesting. But you have to expect nearly everyone is using AI at this point.
It is actually darkly fascinating how eager so many people are about racing towards NPCdom.
Dead internet theory inching closer everyday.
I knew it would be Canada before even opening the link, what a total embarrassment!
Mais par contre, Canada also has some tremendous senatorial committees. They asked real questions of Hinton.
politicians are one of the easiest jobs to replace with AI
We've had the technology to replace what the public percieves as the job of politicians for decades.
I really hope you’re joking, but it’s hard to tell on here!
Indeed. Manual dexterity on a bipedal robot is still in its infancy compared to LLMs. We'll need another breakthrough or two before we have robots that can play golf and schmooze with other rich robots all day.
I feel like the world has lost 30 IQ points over the last two years. I can’t think of a better way to describe it, the dumbing down of everything is palpable and accelerating.
IQ levels are same as before, it's just easier to spot differences.
IQ levels have decreased in the 21st century, it's known as the reverse Flynn effect, Norway did a large scale study on almost a million people because they have a lot of military data and accounting for genetics IQ peaked for people roughly born in the late 70s and has been declining since. We've actually wiped out a fair amount of 20th century gains in intelligence.
Oddly: this coincides with the period in which leaded petrol has largely been phased out on most countries. The use of lead in motor-vehicle fuel is though to have shaved a few points off IQ by itself.
Much as with some other widespread phenomena that come to mind, it's an observed trend the reverse of what might otherwise be expected.
Lead exposure can damage the reproductive system, so perhaps it’s to be expected if a whole generation of parents were exposed to leaded gasoline.
Theoretically this effect would wear off with the next generation, unless there was durable damage to genetics that gets passed on. A cursory search suggests this may be possible, but I’m not a biologist or geneticist.
"Accounting for genetics"?
Oddly, getting spotted for being dumb might be intelligent.
As I click another AI generated ad to tell which company could possibly have so little integrity…
And it had already lost about 30 points through mobile devices and touchscreens, as I've commented before:
See: <https://news.ycombinator.com/item?id=37521911>, also <https://news.ycombinator.com/item?id=42787831> and <https://news.ycombinator.com/item?id=38659367>
For the past several years (since at least 2020, if not a few years earlier), I've commented/posted to HN exclusively from desktop systems.
The bad news is that this is as good as it gets ;-)
* AI Response. Not prompt.
I was expecting him to read a proper prompt. The prompt would be what he would tell the AI system.
“You are a speech writer, an expert on XYZ… your task is to write ABC…”
It’s not too bad, at least he is using ai. And agree we can replace many of them with AI. At least we will have the opportunity to talk to the ai directly about concerns.
One very tiny silver lining in all this will be leaked prompts. E.g. we all know how a non-apology reads, and call them out as such already, but when you see "write a really contrite apology without actually conceding anything while subtly casting doubt on the character of those making the accusations, but in a way that seems really open and constructive" it'll hit different still, I think.
> It’s not too bad, at least he is using ai.
What is not too bad about him using "AI", exactly?
When the yogurt took over
Why exactly is that bad?
They care so little they don't even bother to proofread once, or lack enough intelligence to distinguish a prompt meant for a machine from a speech to inform/inspire/influence a human audience.
It's a red flag. Even if they agree and understand whatever they copied from chatgpt, they were too lazy to clean up its output. Are they also too lazy to read the news or does chatgpt summarize their articles too? AI people are like emperors without clothes. Totally embarrassing to witness.
This is bad because it's only getting harder to trust people. This is not someone I would want representing me. Although I'm not thrilled about any of my representatives anyway.
You're absolutely right. Sure, I'll help you write a Hacker News comment explaining why politicians should have functional mental faculties and a coherent understanding of what they do during the course of their jobs and how their actions affect their constituents.
None of that requires you to be a good public speaker or speech writer
You may have missed the joke that the comment you're responding to is written as if it were an AI response to your original question.
(I'm presuming it, and the other comments of this subthread, are humanly generated. Which may be being generous....)
It was but I'm not sure what in my reply indicates missing it. The comment still made a point and I replied to it