Depends what you mean by 'recently', but for me they are much worse than they were several years ago. There was a period when people were complaining and I didn't really see why, but eventually whatever it was caught up with me too. I think it's a combination of losing the battle with SEO spam and prioritising things other than giving me what I actually ask for. There's lots of obvious junk (either 'AI slop' in the truest sense, or the human-written version that was common pre-AI) that finds its way to or near the top of the results; also, it can be difficult and frustrating to convince Google that I'm actually looking for X rather than the superficially similar and more popular Y, and that I would prefer a small number of actually-relevant results to a million irrelevant or sloppy ones.
I blocked it because I found it was in the sour spot of being good enough to be tempting to rely on, but bad enough to be risky to rely on.
When the search results are bad, usually I can at least tell that they're dubious: either they're from obviously unreputable sites, or they conflict with each other, or they just don't quite address my query. But an inaccurate AI overview can look very similar to an accurate one.
> So no, ”very often works on depression” is not a characterization I would use.
I'm (genuinely) sorry about your friend, and I don't deny that it's worth sharing these anecdotes. But a single anecdote comes nowhere near refuting the claim that ECT very often works on depression.
The current state of scientific knowledge seems to be that it does very often "work", at least as a fast-acting short-term treatment for very severe depression.
> but nobody should be surprised when people vote for an outsider who says "I'm for you, and I'm going to help you take back your country from the out of touch elites who hate you and only look out for themselves"
Sort of, but that was always a pretty obvious tack to take, and I don't think there was ever a shortage of would-be leaders willing to play that role. So we're still left with the question of why the voters chose the most obviously untrustworthy guy to play it.
I think there is a shortage of would-be leaders like that though, that's the problem. Or at least would be leaders that gained any real traction. The only other one in the past decade was Bernie.
Unfortunately for the past 3 elections, it essentially came down to the obviously untrustworthy "outsider" vs the ultimate establishment candidate. For a lot of people, it's as simple as that.
The article explicitly says that the author looked at the diffs; it distinguishes this from "sitting down and actually reading the code", which they didn't do. So when plastic041 says the author spent 7 months vibe coding "without ever looking at source code", it's not unreasonable for dewey to assume that "looking at source code", in this context, actually means something stronger and excludes just looking at the diffs.
Does the original reply actually make sense in context? I can't see how.
It's a response to someone saying "you can't draw any conclusions of IQ significantly before 1950 from how the line behaves after 1950", and it says "And that’s because IQ is a statistical distribution, not an absolute measurement of intelligence."
This seems like a non sequitur to me. Am I missing something? (Bear in mind that the 'line' under discussion is an increase in unstandardised scores.)
On a given set of 1000 questions, over time the trend has been to answer slightly more of them correct every year, progressively raising unstandardized scores, over the set of all IQ testees, since IQ testing was formalized in the 1950s.
Extrapolation is the most questionable statistical tool, and while extrapolation ad absurdum is a way to show a formal predicate logic argument to be incorrect or underspecified, it is an almost fully general attack against real datasets, which basically always have some trend line that ultimately passes sensible thresholds like zero bounds. Showing this, however you form the trend line, is not saying a whole lot.
Extrapolation prior to 1950 is not a very useful tool to evaluate intelligence trends, and this is entirely separate from the periodic recalibration of IQ tests to keep the average at 100 (however many correct answers out of 1000 this corresponds to).
This is another non sequitur ... it doesn't address retsibsi's point or their question. It has nothing to do with cluckindan's comment, which is what this subthread was about.
It's because there are multiple levels of misconceptions as well as "violent agreements".
retsibsi is correct. You can't draw (meaningful) conclusions about IQ before 1950, because extrapolating from the data after 1950 is dumber the farther back you reach, just for reasons related to the concept of extrapolation.
This has nothing to do with the fact that IQ is a statistical distribution that we keep re-norming, which "should always average 100"; The Flynn Effect is not in serious dispute, it's just an effect that pertains to nonstandardized results.
They're not reducible, but I don't know if that means we don't have definitions; we can describe them well enough that most people (who aren't p-zombies or playing the sceptical philosopher role) know pretty well what we mean. All of our definitions have to bottom out somewhere...
> Do insects feel pain?
Nobody (except the insects) can know for sure. Our inability to know whether X is true doesn't imply X is meaningless, though.
> But how can X be a good indicator for something I want to determine if I can’t measure X either?
In the comment that started this subthread, qsera was responding to someone who said "Imo we don't even have a definition of [consciousness]". If qsera meant that we can measure consciousness in terms of pleasure and pain, then of course I agree that they were just pushing the problem back a step. But I don't think that's what they meant.
The person they were responding to said "Open models have the same performance on coding tasks now." AFAIK this is bullshit, but I'd love to be corrected if I'm wrong.
I don't mean this in an "I know better" way, just genuine curiosity: why couldn't you record a solution with pauses and then strip them from the replay file?
But if winning the game requires you to do shitty science and defraud the public, why play it at all? There's no desperation justification here, because anyone who can succeed in academia almost certainly has the brains and credentials to get a decent non-academic job.
Because, for one thing, some people are shitty frauds, and they're not bothered by it. Those people see messed-up incentives as an opportunity.
Do serious workers tend to get out of the field, if the incentives are wrongheaded enough? Sure. Some. Does that fix the incentives or the outcomes within that field? No, not at all.
Because it's not a requirement, and most people are not intentionally or accidentally defrauding the government.
The issue is that there is no incentive to do the additional work necessary to generate reproducible results because of the pressure to constantly generate sufficiently novel results to publish.
If you spend the additional time required to have fully reproducible results and your competition is not, you're probably going to lose the game (where the game is obtaining more funding).
Not generating reproducible results doesn't mean you're a fraud, but the absence of a requirement to generate them in order to publish means that it's easier for fraudsters to operate that it would be with that requirement.
> anyone who can succeed in academia almost certainly has the brains and credentials to get a decent non-academic job.
I suspect the way this usually gets started is similar to embezzlement schemes. “Oh I’ll just borrow a few dollars from the till and pay it back tomorrow” is akin to “The manuscript is due tonight so I’ll just touch up this microphotograph to look like the other one that had bad focus.”
That escalates into forging invoices on the one hand and completely fabricated data on the other. By that point they’re in too deep to stop until they get caught.
Because you've just spent 10-15 years studying for a masters, PhD, and postdoc how to do exactly one thing, and probably are IN that system for another 5-10 years before realizing how totally corrupt it is.
reply