groundsloth

Karma: 179

groundsloth 9 Jul 2026 2:51 UTC
5 points
0 ∶ 0
on: METR Time Horizon 2.0—The benchmark you’ve been waiting for
Plausibly Excel has obtained superintelligence but is, as per the aestivation hypothesis, biding its time and storing energy until the universe cools.

groundsloth 6 Jul 2026 15:41 UTC
−1 points
0 ∶ 0
in reply to: Joel Tan🔸’s comment on: The case for an autonomy prior in global health
Moreover, people generally prefer life to freedom (they wouldn’t choose the death penalty as a punishment over prison or even modern slavery), and if you go out and talk to people in LMICs, you’ll largely find they prioritize the lives and health of their kids and themselves + having a job & escaping poverty over abstract notions of freedom. If you want to centre people’s preferences, you have to take that into account.
I don’t see how this example shows anything. The reason governments have to expend effort enforcing their paternalistic regulations is because these regulations violate the preferences of consumers. Now maybe you think consumer preferences are wrong, but forcibly limiting people’s abilities to take an option that they prefer is clearly not “centering people’s preferences”, unless you think your methodology is somehow better placed to figure out what people’s preferences are than their actual choices.

groundsloth 29 Jun 2026 8:16 UTC
1 point
0 ∶ 0
in reply to: NickLaing’s comment on: NickLaing’s Quick takes
In general it’s okay for a person to look at a dozen different paintings and then make a new painting that’s kind of like those paintings. This seems pretty analogous to what AI is doing and I’m not sure why it becomes not OK if it’s done by a corporation training an AI model. Perhaps there are specific violations of IP laws, and those can be discussed (and some of them are being adjudicated in court as we speak), and of course there is a separate question of whether those IP laws are just. However, what AI models doing to me seems mostly like it’s the sort of behavior we would be OK with individual people doing: i.e closer to the remixing/synthesizing end of the spectrum than the copying/”stealing” end.

groundsloth 29 Jun 2026 8:01 UTC
7 points
0 ∶ 0
in reply to: Benevolent_Rain’s comment on: Ulrik Horn’s Quick takes
https://www.interceptfund.com/

groundsloth 22 Jan 2026 5:14 UTC
3 points
1 ∶ 0
in reply to: Robi Rahman🔸’s comment on: Propose a debate topic—I’ll run one
I don’t know how philosophically sound they are, but the following rules, taken from the RP moral parliament tool, would end up splitting donations among multiple causes:
- Maximize Minimum; “Sometimes termed the ‘Rawlsian Social Welfare Function’, this method maximizes the payoff for the least-satisfied worldview. This method treats utilities for all worldviews as if they fall on the same scale, despite the fact that some worldviews see more avenues for value than others. The number of parliamentarians assigned to each worldview doesn’t matter because the least satisfied parliamentarian is decisive.”
- Moral Marketplace: “This method gives each parliamentarian a slice of the budget to allocate as they each see fit, then combines each’s chosen allocation into one shared portfolio. This process is relatively insensitive to considerations of decreasing cost-effectiveness. For more formal details, see this paper.”
There are a few other other voting/bargaining style views they have that can also lead to splitting.
I don’t really have anything intelligent to say about whether or not it makes sense to apply these rules for individual donations, or whether these rules make sense at all, but I thought they were worth mentioning.

groundsloth 20 Jan 2026 21:27 UTC
4 points
1 ∶ 0
in reply to: Robi Rahman🔸’s comment on: Propose a debate topic—I’ll run one
You seem to be assuming a maximize-expected-choiceworthiness or a my-favorite-theory rule for dealing with moral uncertainty. There are other plausible rules, such as a moral parliament model, which could endorse splitting.

groundsloth 20 Jan 2026 16:29 UTC
2 points
0 ∶ 0
in reply to: Robi Rahman🔸’s comment on: Propose a debate topic—I’ll run one
Why would this be? For example, could not an individual donor be uncertain of the moral status of animals and therefore morally uncertain about the relative value of donations to an animal welfare charity compared to a human welfare one?

groundsloth 13 Dec 2025 20:10 UTC
3 points
0 ∶ 0
in reply to: David T’s comment on: Two arguments against patient philanthropy
my example was merely trying to show that if patient philanthropy is preferred, then a wait and donate strategy is better than relying on inter-generational transfers.
we have to assume that for some arbitrary reason we do not care about the first three generations of poor people at all, only about the poor people in 100 years’ time
No, we don’t. We merely have to believe (in expectation) that our marginal money is better spent a few generations in the future than on the current generation. This is of course contested but there are plenty of non-arbitrary reasons to believe it. For example, if you doubt catch-up growth, and you think there will still be some very poor (in absolute terms) countries around a few generations in the future, then a few generations the future you will expect to have
- more money, due to compound returns of your savings
- a population of people that are about as easy to help as they are now
So you can help people more then than you can now. (this is obviously glossing over a ton of details like those covered in Trammel’s report but I think it helps get at the intuition).
Most individual-level income gains like we see from tin roofs do not compound across generations at anything like market rates (we can see inter-generational income elasticity is far less than 1 in low-income countries which implies income gains to one generation get diluted over time).

groundsloth 12 Dec 2025 3:27 UTC
20 points
5 ∶ 0
in reply to: Yarrow Bouchard 🔸’s comment on: Two arguments against patient philanthropy
So lets say we are targeting population P—these are the recipients. P is the population of people most in need that will be around in 100 years. Most of them do not currently exist. We want to spend money to help P either in the form of direct cash transfers or health interventions.
We can do that by investing our money and then handing it out to the individuals in P once they come into existence. This is option A, aka the patient philanthropy strategy.
We could also give to the parents or grandparents of P, some of which are alive today, which we can call P_p and P_g. I am assuming this is what you mean by “Give it to their ancestors.”. I call this Option B. Option B is worse because:
- some of the money is consumed by P_p and P_g instead of passed thru to P. In this example we are assuming P is where our money is the most cost-effective so this is bad.
- since we don’t know what areas will be most in need for the next few generations we don’t know quite who P_p and P_g are.

groundsloth 12 Dec 2025 3:14 UTC
8 points
2 ∶ 0
in reply to: Yarrow Bouchard 🔸’s comment on: Two arguments against patient philanthropy
see my previous comment:
- Option A: Put money in an index fund, let it grow, spend it in 100 years
- Option B: give it to people who will spend some of it now, invest some, pass on some to their children, who will in turn spend some of it, invest some of it, and pass it onto their children, leaving some for their grandchildren to spend in 100 years.
Option A leads to a bigger counterfactual increase in spending-100-years-from-now which is what we care about in this (admittedly contrived) example
Giving it to their ancestors is choosing option B

groundsloth 12 Dec 2025 1:11 UTC
8 points
2 ∶ 0
in reply to: Yarrow Bouchard 🔸’s comment on: Two arguments against patient philanthropy
Well in this case the recipients do not currently exist so I cannot yet give them any money.

groundsloth 12 Dec 2025 0:48 UTC
8 points
0 ∶ 0
in reply to: Yarrow Bouchard 🔸’s comment on: Two arguments against patient philanthropy
Not necessarily. Let’s say a patient philanthropy foundation wants to expropriate, say, half of the inherited wealth between this generation and the next (in order to invest it for a future generation). The current generation would rationally prefer this not to happen. Conversely, the current generation would prefer to receive from the foundation an amount of wealth equivalent to half the wealth the next generation will be able to inherit.
The fundamental point is that investing wealth to accumulate wealth might make your impact numbers go up, because of an arbitrarily accounting decision, but it doesn’t make the well-being of the people you’re trying to help go up, as measured by their rational preferences or revealed preferences.
I’m not sure I understand this line of argument. Let’s say the world is such that spending in 100 years (i.e on people who do not currently exist and cannot actualize their preferences) is especially cost-effective or otherwise useful. There are two ways of “putting money 100 years into the future”.
- Option A: Put money in an index fund, let it grow, spend it in 100 years
- Option B: give it to people who will spend some of it now, invest some, pass on some to their children, who will in turn spend some of it, invest some of it, and pass it onto their children, leaving some for their grandchildren to spend in 100 years.
Option A leads to a bigger counterfactual increase in spending-100-years-from-now which is what we care about in this (admittedly contrived) example.
I agree setting up 100-year foundations is hard; but there are more practical steps one can take if they are convinced of the general arguments for patient philanthropy, most simply giving later in their lives or in their wills (I’m not necessarily endorsing this, but I think it is worth considering).

groundsloth 11 Dec 2025 23:54 UTC
24 points
4 ∶ 0
on: Two arguments against patient philanthropy
The rational preference argument only applies for giving to the current generation of recipients at some later point in their lives. If the most cost effective generation to help is really, say, 3 generations in the future we should save and then give to them instead.
The cost-effectiveness argument is more convincing but the case for patient philanthropy only requires there to be some large-enough region that stagnates growth-wise which unfortunately seems likely given the perennial nature of bad governance. I believe that if you make some simplifying assumptions about utility functions and population size, patience looks good iff the projected rate of return on your assets is greater than the projected income growth rate for your target population.
I agree donating now to promote economic growth in poor countries is likely preferable to delaying donations by 100 years, though I am unsure about the tractability of any particular growth intervention.

groundsloth 3 Dec 2025 5:53 UTC
6 points
2 ∶ 0
in reply to: Larks’s comment on: Epistemic Humility vs. Tobacco Abolition
I also found their neglect to mention to mention smokeless products strange. In Norway and Sweden, for example, snus has been replacing cigarettes.

groundsloth 3 Dec 2025 5:32 UTC
7 points
0 ∶ 0
in reply to: Jason’s comment on: Epistemic Humility vs. Tobacco Abolition
There are not too many case studies to look at here, but in Bhutan, their 2010 ban on the importation and sale of all tobacco products (which was coupled with government counseling and treatment to facilitate cessation) proved controversial and difficult to enforce in the face of large-scale smuggling and was eventually scaled back in 2021.

groundsloth 1 Dec 2025 17:11 UTC
15 points
3 ∶ 2
in reply to: titotal’s comment on: Epistemic Humility vs. Tobacco Abolition
fair points! I should have been explicit here but I was invoking epistemological humility primarily in response to SMA’s public statements about the tobacco industry which in my view do not seriously engage with the idea that tobacco use could be a free-but-harmful or free-and-beneficial choice for any consumers.
I would push back on the idea that the harm/benefit ratio to smoking is somehow ‘medically settled’. The fact that cigarettes greatly increase risk of lung cancer, say, is settled but for the harm/benefit ratio to be settled it would require us to be able to put the harm and benefit in comparable units; the correct procedure for which is, in my view, a epistemological mystery (the problem here seems similar to trying to get an unambiguous answer with health economics whether to go on a fun-but-risky rafting trip or a safe-but-boring trip to the mall).
I do believe the freedom to choose self-destructive habits is an important principle worth defending though I don’t think that claim is necessary for the more narrow point I was making in this post.

Epistemic Humility vs. Tobacco Abolition

groundsloth1 Dec 2025 9:15 UTC

42 points

19 comments3 min readEA link

groundsloth 10 Nov 2025 18:21 UTC
1 point
0 ∶ 0
in reply to: Buck’s comment on: Lack of Diversity Within EA
In what way do you find it unrepresentative? Just curious because I am unfamiliar with the dynamics here.

groundsloth 25 Sep 2025 5:26 UTC
6 points
0 ∶ 0
in reply to: tobycrisford 🔸’s comment on: A Simple Case for Patient Philanthropy
good points!
with respect to the utility function I originally chose log because it’s what Open Philanthropy uses. However I now see that GiveWell sometimes uses an isoelastic utility function with $η = 1.59$ , which is faster diminishing returns than log-utility ( $η = 1$ ).
Redoing the math with a general isoelastic utility function, for $η > 1$ you get an optimal point at which to donate, with that point depending on the parameters of the model (income and investment growth rates, the rate of diminishing returns, etc.). This optimal donation time is also dependent on the size of the population you can donate too, so to get a more accurate model you would need to incorporate that (as well as a bunch of currently-unaccounted-for factors like the changing effectiveness of non-cash charities).

groundsloth 23 Sep 2025 22:56 UTC
2 points
1 ∶ 0
in reply to: AgentMa🔸’s comment on: A Simple Case for Patient Philanthropy
I think value drift (along with risk of losing control of your assets due to any number of scenarios) can just be modeled by discounting the expected growth rate for risk of loss. If you think you have a 1% chance of losing your money each year, you might treat your investment growth rate as 4.2% instead of 5.2%, for example.

groundsloth

Epistemic Hu­mil­ity vs. Tobacco Abolition

Epistemic Humility vs. Tobacco Abolition