(Careful! This statistic is not useful in predicting if the total number of any species has increased or decreased. The data could say all species had populations that massively grew, and it would still report a decline. See Our world in data for a toy example, and discussion of what the base-data are for LPI.)
Ok; I’ve thought through several examples. The geometric mean is heavily biased to show a decrease IMO. A species could have it’s total population grow substantially, but if most of that growth is to already large/successful subsets, then it’ll almost certainly show as a negative (geometric) average change, even with ~noise on the other populations. I think you should only see a positive metric here if most species measured are invasive; growing extremely fast in places not prepared for them. (and being careful not to measure the things getting killed by the new invaders)
Broader than capitalism; it’s growth driven no? A planned economy that wanted a steady increase in energy/quality of life/resource extraction would also do most of this.
I don’t think starvation due to ecosystem collapse is widely expected anytime soon. That would be the headline otherwise, right? In particular, we pretty much control the whole grain + maize + soy + rice production process afaict?
This is a famously difficult to interpret statistic. It’s an excellent survey, but you probably do not understand what the number means (because it is a weird number). In particular, only half of species surveyed showed a population decline at all, the other half either grew or were stagnant.
The authors of the LPI go to great lengths to explain this. Yet, there is inevitably a long list of headlines claiming that we’ve “lost 73% of wildlife”, “73% of species have gone extinct”, or “73% of species are declining”. All of these are incorrect.
To give you another sense that this number is weird:
The 2022 Living Planet Index reported an average decline of 69% since 1970. In 2024, it reported a 73% decline. And this latest report finds the same: 73%.
So, what exactly is this 73% number?
To obtain the final LPI number, researchers take the geometric mean of the rates of change across all these populations. They also provide different weights to different populations, which affects the final result.
I don’t know other places where I’d take the geometric mean. I have no intuition for it. So… caution advised.
EFF is competent, and there’s plenty of legitimate open source projects with reasonable name recognition. Also, like, android (google) and vscode (ms). Plenty of headline products that are open source and want to keep running. I think there’s interests on both sides of this one.
Could you walk me through the damage that this mistake does to well-and-proper open source developers?
I don’t think it is likely that they would be harmed by things like new regulation on ‘public source code’ projects? Anybody looking to hire a developer who confuses these two wasn’t serious/relevant in the space. Open source developers do not, at least in my media diet, seem to get a lot of airtime or respect for their work (or money, for that matter). And the places they do get these things are sufficiently literate that I’m not concerned.
Could you help me out with this claim? What’s the model of harm?
Great; that describes all functions, so we’ve nothing to worry about.
Cool; walk me through attention layers plz.
to be fair, shutting down yourself is pretty easy on the corporate espionage scale.
I recommend politics. Local politics can get their presence in your neighborhood to chill, set examples for other municipalities, and lead to broader office (if done well). Legal, community driven, and fairly direct. Does require talking to your neighbors, sometimes in public, which is a substantial downside.
Failing politics, I think your next best bet is probably attempting to do some hacking/espionage work. LLMs make this more scalable than it used to be, and they seem pretty happy to hack big companies. Costs a lot of money, will enrich other tech bros with high probability, definitely illegal, and it’s pretty hard to do this hard enough to crush the business (tho if you can siphon enough money out???). But you won’t have to talk to a neighbor or speak in public, so that’s nice.
Isn’t the saying “hate the game, love the players”.
Ty for the show recommendation!
I am pretty familiar with much of the (mathematics) story for Newton and Euler; while they did accelerate things, famously much of calculus from Newton would have emerged from others within a decade or two. And it’s not uncommon to be working anywhere in mathematics and stumble upon a ‘new to me’ result which is, inevitably, really due to Euler. This isn’t to say they did bad work, just that the large part of their work was available by paths with few ‘breakthroughs’.
I believe several of these bots run with APM limits; so they also skip some of these.
(Should be called ‘genie’ problem instead of hacking/cheating. Tis a cursed monkey’s paw we’ve created.)
While there’s some of both, I dislike the ‘great white man’ narrative of breakthroughs in science. There have been legitimate breakthroughs, but most (maybe all) of them were inevitable from recombination and mild improvements.
(A reminder that wishing your political enemies die by their own hand is… not great.)
GenAI is an averager (as all empirical models are), so it tends towards predicting the mean of whatever it’s trained on (in a given context).
Perhaps this sentence is not what you wanted then? The mean of whatever it’s trained on (that is, the training data) need not be the mean of the trained LLM function. This is because we use complicated loss functions to update our LLMs, stochastic processes in training, and that we change those functions after training on input data by using RLHF and other models (eg, via using a mixture of itself to get better outputs with the boosting algorithm).
You are right that once we fix an LLM’s parameters, at temp 0 they are deterministic, though they are recursive functions so it’s weird to talk about a long string of outputs at temp 0 as the ‘mean’ of the LLM; high temp outputs won’t ‘regress’ to this temp 0 output over time, for example. There also may be hallucinations or mistakes that are common at temp 0, but extremely rare at all other temperatures (because these mistakes only get made when it looks at its own temp 0 context). So I’m not sure the deterministic output string(s) are particularly useful.
(An analogy that I suspect isn’t very good: the butterfly effect, adding a single small air perturbation can lead to drastically different deterministic weather simulation results. This is very much the norm for LLM outputs)
(They are somewhat more than a neural network; the ‘attention’ layers add something pretty strange. And output temperature isn’t implemented as a noise term, but this is a fine way to think of it.)
That’s correct; dictators rarely understand how the weapons they use work. I doubt there are many politicians who can describe how a nuke works, but they’ll continue to threaten to launch them. This is not a bold or surprising claim, is it? All they need to worry about is that it works ‘often enough, at a cheap price’.
And my point is not that you should use AI for addition. We agree it is the wrong tool (and modern models will just call python). My point is that AI completes tasks in weird ways that are hard to predict; the sweatshop analogy, remember?
Finally: Maths is not magic handwaving, but there are many many ways to perform simple addition. Homework exercise for those at home, find 3 different (equally correct) algorithms to add single digit numbers.
Managing employees is miserable, eh?