So I don’t think that is true. It’s possible to recognise that a book is well written even if you can’t write that well.
I think the problems from LLM use in that area are more about hallucination, if it inserts a false job or something, which is easily checked. OTOH if it just edits it and it looks no better to your eyes, you’re probably ok to go with your initial version.
What does an error in a CV look like, to you?
I’m not really here to tell you why you should care - you’re free to care about whatever you want to care about. But to explain why other people might care, it’s because it can do things a Google search can’t do. Google search can’t copy-edit your CV or cover letter. Google search can’t synthesise a bunch of different Stackoverflow answers and fit them to the exact scenario you’re talking about. LLMs can and do.
And those are two examples where the cost of an error is low: if your CV comes out with made up shit in it, you can just read through it and check (but you may not have the ability to re-write it better). If the code example doesn’t work, you’re going to run it and check anyway. (It may have a subtle bug, but so can Stackoverflow answers, and that never stopped people from using them)
Non-deterministic software is fine and we’ve been using it for ages. It’s usable when:
That rules out several applications of current LLMs, but it rules in several others.
It’s about 10x more power intensive than a Google search. It’s not trivial, but it doesn’t take megawatts to power a single person’s query.
I tend to agree.
But because the system prompt is part of the context, it figures into the estimation of the most likely next token. So in general putting this kind of stuff in the system prompt does change how well it works.
What matters is the outcome, not how it is achieved.
And is the outcome good? Eh, sometimes.
I see
“Why is it OK to use a metric” is a weird question to ask. Why wouldn’t it be OK?
You could certainly try and take other things into account but… do you believe that the UK government does not consider any metrics except GDP when designing policy? Do you believe that voters don’t consider any other outcomes when deciding whom to vote for? These are clearly not the case, and using several metrics together is equivalent to tweaking one metric to incorporate additional facts.
The way we really see the particular example you picked is by comparing unpaid and paid childcare. So the government could subtract paid childcare from headline GDP statistics, on the assumption that what that actually pays for is something that would otherwise happen anyway without payment.
But what would making this tweak achieve? Do you think people are out there pushing for an expansion to free childcare because it would make GDP figures go up? Because I think we push for that because it’s the right thing to do for society.
As for my example, I think if you thought a little bit you’d see how practical ways of measuring depression (such as numbers of diagnoses in the NHS) would be subject to the same perverse incentives you’re talking about with how GDP is measured. You suggested some proxies for wellbeing. Let’s take number of sick days: the perverse incentive there would be that the government might launch a crackdown on slackers taking sick days they don’t need to make the numbers look better, even though that’s bad for society.
Maybe this isn’t the kind of thing you’re worried about with GDP, but then I don’t really knwo what is.
It’s called driver config and it’s not a tremendously complicated bit of software. Most of its functionality doesn’t need to “inject into games” - it just needs to remap the inputs so the driver can present them in whatever way works for the game. If the game doesn’t support anything sensible, then maybe fancy stuff is warranted, and you might well miss out on that from a driver config utility. But that’s OK.
The simple answer is because, while it’s not perfect, it’s still quite useful, and easier to measure than something more subjective like wellbeing.
Consider an analogous satire to yours: “If you’re depressed, or demanding better conditions, you are ‘harming’ the country’s wellbeing - you have reduced your wellbeing and are not granting wellbeing to anyone else by doing so. Why don’t you lower your expectations and be happy with what you’ve got, why don’t you avoid seeking treatment for depression so it’s not recorded on the statistics!”
All such metrics which aim to comprehensively summarise the state of something complex like an economy or population are subject to things they can’t capture, and subject to perverse incentives.
I think a good, honest attempt to make wellbeing a metric by which we judge the country is a fantastic idea, but a simplistic take on GDP - like yours, I’m afraid - is not why.
That’s for laptop keyboards - I don’t see any spare parts there for membrane keyboards. I said “rubber dome” but I guess that was ambiguous; I meant the PC keyboards where there’s a single moulded rubber sheet inside that forms the switch and “spring”. I was not able to quickly find anywhere offering spare membranes for ordinary keyboards, but I’m pretty sure they’ll be more than 5 euro :P
Dick down December
Sissy September
Norks November
Jelly July
Milf march
A keycap is that? I don’t believe that is how rubber dome keyboards break… I think the rubber sheet degrades so that it no longer closes the switch when pressed, and you need to open up the whole thing, clean it, and replace the whole sheet (which is custom for each board, not as generic as a switch) if you can find one.
Most keyboards (not in laptops) are rubber dome ones though… Significantly less economical to repair.
Ragebait or just being a shit person? Get in the bin either way, I don’t need to see more from you.
It’s a joke.
This can be the case for coding. A good example is when the change is simple but involves a library you’re unfamiliar with. You can set it off and not have to read any docs, and it will be easy to check if it got the API right.
Elsewhere I gave the example of copyediting. It’s a lot quicker to check the output than to refine it yourself.
Easy-to-verify tasks are everywhere I think. Not at the scale of seconds versus hours, but seconds versus minutes