RHollerith

Karma: 2,994

Richard Hollerith. 15 miles north of San Francisco. hruvulum@gmail.com

My probability that AI research will end all human life is .92. It went up drastically when Eliezer started going public with his pessimistic assessment in April 2022. Till then my confidence in MIRI (and knowing that MIRI has enough funding to employ many researchers) was keeping my probability down to about .4. (I am glad I found out about Eliezer’s assessment.)

Currently I am willing to meet with almost anyone on the subject of AI extinction risk.

Last updated 26 Sep 2023.

RHollerith 1 May 2026 19:58 UTC
2 points
0
in reply to: lilkim2025’s comment on: AI unemployment and AI extinction are often the same

power concentration were in the extreme, and one person controlled all of the LLM instances doing all of the jobs

Some of us would consider that a good outcome (relative to what we think is likely to happen) because at least humanity does not go extinct (and Carl Shulman made that point on this site already in about 2013). We just consider it unlikely that any person retains enough control to keep even himself and his friends alive as AI becomes sufficiently capable.

To be precise, it is not strictly necessary for any person to retain any degree of control. The crisper way to say it is that for any part of humanity to survive, AI (i.e., all the models considered as a system that has some effect on the world) must care at least a tiny bit about what at least one person wants, but sadly this property of caring at least a tiny bit that IMHO is unlikely to be satisfied—because no one knows how to create an AI with that property and no one is likely to figure out how to in time. We started calling it “AI alignment” about 12 years ago (before “alignment” came to mean “corporate brand safety”) but we could’ve called it “AI caring” or “AI inter-species regard”.

I believe that what I just wrote applies whether the AIs “win out in one crushing step, or win out in a trillion small familiar ways,” to quote from Katja’s final sentence.

RHollerith 25 Apr 2026 17:29 UTC
4 points
2
in reply to: streawkceur’s comment on: Do not be surprised if LessWrong gets hacked
I strongly urge Lesswrong to resist calls to respond to deletion requests by routinely deleting things that have already been published. (I’m OK with deletion of a user name from the public LW archive.)

Lesswrong is unusual among online forums in how much time and mental energy it puts into reflecting on past public writings on the site. The annual review for example invites us to discuss posts the youngest of which are 11 months old.

I routinely copy URLs into text files on my computer so that I might return to them later. A quick search found 37 URLs that begin lesswrong.com/posts/ and 26 URLs that begin lesswrong.com/lw/. Picking one of the 37 URLs, I find it leads to a post made in 2018. Picking one of the 26 URLs, I find it leads to a comment made in 2010. (Part of the comment goes, <<Agreed. I’m often somewhat embarrassed to mention SIAI’s full name, or the Singularity Summit, because of the term “singularity” which, in many people’s minds—to some extent including my own—is a red flag for “crazy”.>>)

There has been a complete change in the software behind lesswrong.com between 2010 and now, but some programmer took the time to ensure that the URLs that worked on the old software would continue to work on the new software. A huge thank you whoever did that work! LW’s efforts to keep old URLs working have made me significantly more effective at reflecting and deliberating on the kinds of things that LW talks about.

Please don’t routinely or automatically grant deletion requests. That would make LW worse at helping people deliberate on timescales of years (e.g., revisiting a piece of writing after a number of years) which is important.

RHollerith 25 Apr 2026 12:08 UTC
2 points
0
in reply to: robert_horn’s comment on: Some takes on UV & cancer

UV light is raising your testosterone levels (much?) less

I doubt my doctor chose a dose that would average out to be much higher than the amount produced endogenously by a healthy young man. (I was young at the time.) My guess is that the exogenous testosterone was causing testosterone levels to vary in a pattern different from the pattern caused by endogenous testosterone production. For example, endogenous production might result in a much steadier concentration.

UV light proably has significant positive effects other than tending to raise testosterone. Alexis Cowan asserts that it is a more potent stimulus to endorphin production than aerobic exercise is.

RHollerith 24 Apr 2026 18:09 UTC
3 points
2
on: Your Supplies Probably Won’t Be Stolen in a Disaster

risk is mostly external, homes are rarely targeted

It is worth considering why this might be. Before committing themselves to a course of action that might turn out disastrously, it is much more difficult for a marauding gang to assess the fighting strength of the inhabitants of a structure (unless they manage to get a spy into the structure somehow) than to assess the fighting strength of a travelling party especially if the party is travelling on foot or exiting or entering a vehicle.

RHollerith 23 Apr 2026 0:32 UTC
2 points
0
in reply to: Pedro Afonso’s comment on: Reflectively stable consequentialists are expected utility maximisers
Be patient. A post like this one requires most of us hours of mental labor to understand well enough to form an opinion on whether LW should have more posts like it or fewer posts like it (the standard we are encouraged to use for upvotes and downvotes).

RHollerith 15 Apr 2026 7:01 UTC
2 points
0
in reply to: HedonicEscalator’s comment on: Some takes on UV & cancer
Thank you for that. I tentatively accept your assertion that the British study (summarized by the Youtube video I referenced) is not credible.

For science, I wrapped my body in a polyester blanket and wore a sun hat with UPF of 50+ and sunglasses so that none of my skin was getting UV at any significant intensity and spent half an hour in the mid-day sun. All sunglasses on the market from name brands reliably block UV. Polyester is very effective at blocking UV and like almost all fabrics will let at least half of the photons at most infrared wavelengths through. I did this prol at least 20 times. This protocol might have had some positive effect, but if so, the effect went unnoticed by the methods I used (namely, mere introspection and curiosity) or the effect was not large enough for me to be sure that the effect is actually there. In contrast, 20 minutes of mid-day sunlight on a cloudless day on my bare skin and into my “bare” eyes has a pronounced positive effect which are obvious to me.

(Also I once had a session in a machine that costs many 1000s of dollars that I climbed into naked that bathed me in all directions with red light and near-infrared light produced by thousands of LEDs. I did not notice the effects that I value the most from sunlight exposure.)

That strongly suggests that some combination of visible light and UV light is the cause of the effects I value the most. Here let us observe that the common LED light bulb produces no UV light.

I’ve never made the experiment as far as I can recall of exposing myself to LED light bulb light that is as intense in the visible spectrum as the mid-day sun is. (Frankly I would probably strongly dislike the experience.) But I definitely notice effects on my internal experience from the brighter variants of the ordinary LED bulbs that most of us have in our homes, and the quality of those effects differ drastically from the quality of the effects from sunlight, which causes me to conclude that the UV fraction of sunlight is in expectation the cause of most of the positive benefit I get from sunlight exposure.

Specifically, within only a few seconds of the start of exposure to bright LED light, I am more alert, maybe it is easier for me to pay a certain kind of attention or focus and on a minute-by-minute basis, it becomes easier (requires less willpower) to turn intentions into sequences of actions. Sunlight has the same (almost instantaneous) effect, but also has other effects that LED light bulb light does not cause. In particular, the effect I value the most persists for hours after the end of the exposure to the sunlight and might persist well into the next day. A kind of optimism and bias for action, but very steady. also, the quality of my cognition seems to be better (in hard to describe ways) on a day when I get 20 or 30 minutes of late-morning or mid-day sunlight and use only the intensity of LED light needed to see comfortably than on a day when I get little or no sunlight, but spend the day under LED light of the intensity one would find in an average office or store (i.e., much more LED light than I tend to use when I am in my home).

There is a lot more I could say on the subject (e.g., on mechanisms) but the essence is that there is a fuzzy argument relying heavily on subjective evidence for benefits (mainly mental) of high expected value from my moderate UV exposure and a very crisp argument, the soundness of which I do not doubt at all, compiled by dermatology researchers and ophthamology researchers, for pernicious harms (namely, melanoma and horrible eye diseases) of low expected magnitude, and I judge that the former easily trumps the latter because I trust my ability to evaluate fuzzy arguments (i.e., to avoid being systematically biased by, e.g., motivated stopping, motivated continuing or confirmation bias).

The expected benefit of my UV exposure is high because it is almost certain that various (mostly mental) benefits are being manifested and because even though I cannot be precise in describing those benefits or in evaluating the (personal) utility of those benefits, the utility is quite significant in magnitude. The expected harm of my UV exposure is low in comparison because the probability of the most pernicious harms (melanomia, etc) is (as you note) low.

RHollerith 14 Apr 2026 18:52 UTC
2 points
0
on: Americans For Moskovitz
The big advantage we have is that the members of the elite and their families are just as vulnerable to AI extinction risk as the rest of us are. This is in sharp contrast to most deliberations on policy in which the deliberators are reduced to putting hope in the altruistic impulses of an elite who are mostly impervious to the consequences of the policy decision. As soon as any significant fraction of the power elite form an accurate understanding of the danger their families are in, we can expect very drastic moves to curtail the risk. The big disadvantage we have is time: look how long it took for the political will to address global warming to form in Western Europe, and AI extinction risk is probably much harder to come to an accurate understanding of than global warming is.

RHollerith 10 Apr 2026 2:27 UTC
17 points
0
on: Some takes on UV & cancer
https://www.youtube.com/watch?v=sYnUPxG7ONk—Here is a doctor with a Youtube channel discussing a recent British study showing benefits from UV exposure: “if everyone in the 419K cohort avoided UV light and were followed for 15 years, it would save 39 deaths from melanoma at the expense or 2,982 deaths from everything else”.

https://pmc.ncbi.nlm.nih.gov/articles/PMC8411113/ -- UV exposure (namely, 20 to 30 minutes of sunlight exposure without sunglasses or sunscreen every other day) increases levels of the sex hormones in men and women. For men, this increases motivation, drive and interest in sex and romance. For women, it increases interest in sex and romance.

I started routinely getting 20 to 30 min of mid-day sunlight every other day on my bare skin with no sunscreen at all and in my eyes (with nothing in front of my eyes) about 17 months ago. It has positive effects on my mood and motivation and I intend to keep doing it. I’ve also in the past supplemented testosterone, both via oral administration and via injections. I much prefer getting sunlight: exogenous testosterone was hard on me somehow (on my liver if I had to guess) and I experienced negative behavioral effects such as suddenly becoming much more interpersonally aggressive than the situation calls for. It was a bother as much as a benefit whereas I see the sunlight exposure as pure benefit except for the time expenditure.

I remain as careful as ever of not getting sunburned. Part of that is routinely measuring the duration of my UV exposure, and learning some stuff to help estimate UV intensity. I calculated for example that at my latitude (Bay Area) on a cloudless day, the intensity of UV A at noon mid-winter is about half that of noon mid-summer.

RHollerith 6 Apr 2026 17:35 UTC
4 points
0
in reply to: Thomas Kwa’s comment on: Max H’s Shortform
Decoys would not be a problem for the US. A gamma-ray spectroscope weighs only a few ounces and costs only a few thousand dollars. It is almost certain that no one can produce a substance that looks like U-235 to a gamma-ray spectroscope that is cheaper to produce that highly-enriched uranium.

I don’t dispute your larger point.

RHollerith 6 Apr 2026 16:34 UTC
2 points
0
in reply to: Max H’s comment on: Max H’s Shortform
1979 can be seen as a return to traditional Iranian governmental policies after an experiment of a few decades with Western policies. Yes, the form of the current government (namely, a republic at least in principle) is more modern and Western than the form of the pre-1979 government (namely, a heriditary monarchy), but the monarch’s policies were more secular and Western than the current regime’s policies.

Islam has always been political in Iran, which has never had anything like the West’s separation between church and state except maybe to some extent during the experiment with Western policies that ended in 1979. We in the West shouldn’t attack countries just for being different from the West.

RHollerith 6 Apr 2026 15:48 UTC
2 points
0
in reply to: Max H’s comment on: Max H’s Shortform

Toppling a fanatical Shia Islamist regime would be an extremely good thing.

I’m not convinced that it is necessary to topple it. Iran has been Shia for 5 centuries, and during those 5 centuries has attacked its neighbors very little.

RHollerith 3 Apr 2026 18:30 UTC
2 points
0
in reply to: RHollerith’s comment on: Buck’s Shortform
Another reason I trust Deep Mind more than the others is that Gemini lags Claude and OpenAI’s services in coding skill, a dangerous capability because over some (unknown) threshold of coding skill, a model will tend to become capable of effective recursive self-improvement.

I could easily change my belief here though especially by my getting more information about Deep Mind.

RHollerith 28 Mar 2026 2:41 UTC
2 points
0
in reply to: Lukas Finnveden’s comment on: Buck’s Shortform
I accept the correction. I wish there were some other LLM-based service I could recommend to friends, ideally one run by people who would prefer a decades-long pause on AI “progress” and who are only in the market to deprive the other services of revenue and to limit how much the other services can leverage their position as trusted sources of information to influence public opinion on the dangers of AI “progress”.

RHollerith 27 Mar 2026 21:46 UTC
6 points
−7
in reply to: Buck’s comment on: Buck’s Shortform
I picked Gemini as the service to recommend to friends (and to use for my own needs) because it is the only one whose leader publicly committed to a pause if the other leading labs pause and because it introduced a competitive LLM service (Gemini) only in response to other two labs’ getting traction in the market for LLM services, which is a sign (admittedly a weak one in itself) that they were delaying in bringing state-of-the-art AI to market.

IMHO, the single most effective thing anyone can do is to slow down “innovation” in AI since that gives humanity more time to try to solve super-alignment (and IMHO work on super-alignment can proceed effectively without access to powerful models).

RHollerith 20 Mar 2026 0:19 UTC
1 point
0
in reply to: sam’s comment on: sam’s Shortform
A prominent psychologist maintains that if a person is still ruminating about something that happened over 18 months ago, that is a sign that he or she should seek help or at least intensify his or her deliberate efforts to understand the personal implications of the disturbing event or disturbing realization. And sure enough, almost exactly 18 months after Eliezer’s (2022) April’s Fools post, which was a huge update for me, the knowledge that humanity is probably not going to survive the AI juggernaut stopped feeling raw and started feeling like an ordinary fact about the world.

Maybe I was better prepared for that shock than most people are because I’ve experienced a lot of trauma and adversity?

RHollerith 6 Mar 2026 3:11 UTC
3 points
1
in reply to: TsviBT’s comment on: TsviBT’s Shortform
The objections River made apply to the thing you linked, too: namely to stay in a low-earth trajectory for any significant fraction of one orbit requires a speed of 28,000 km/h and more importantly all of that speed must be tangential (“horizontal”). It is expensive in energy to get rid of that tangential component of momentum, and most of it must be gotten rid of in order for the warhead to intersect the Earth’s surface with any accuracy.

(Yes, ICBM’s reach that speed, too, or close to it, but only when the direction of travel is close to straight down. I.e., the tangential component of velocity never gets above a few 1000 km/h.)

Yes, it gets a lot cheaper to get rid of speed when the vehicle is designed to interact with the atmosphere like the Space Shuttle was, as opposed to just shooting through it like a bullet or an ICBM warhead is, but that does not support your point (Tsvi) because such vehicles are the subject of intense study by all the advanced militaries (under the name “hypersonic glide vehicle”) and I have seen no signs that any nation is willing to forswear investment in or deployment of this new class of weapons.

In the decades during which hypersonic glide vehicles were infeasible, River is probably correct in asserting that there was no military advantage to be got from either nukes on satellites or fractional orbital bombardment systems.

RHollerith 5 Mar 2026 17:37 UTC
3 points
0
in reply to: Ebenezer Dukakis’s comment on: TsviBT’s Shortform

is concerned that Russia is seeking to deploy nuclear weapons in space

This was forseeable when SpaceX decided to takes sides in the Ukraine conflict since Russia does not have (and probably cannot afford to create) a constellation of anything like the number of satellites in the Starlink constellation.

RHollerith 16 Feb 2026 21:40 UTC
4 points
0
in reply to: Garrett Baker’s comment on: Linch’s Shortform
Scientists worry a lot that their work will be ignored, which causes them to immitate the style and form of previous impactful works, which explains Newton’s choice of Latin and his choice of proofs in the geometric tradition to present results that he came up with using more calculational and more analytic methods. When I read Principia I felt the author was trying to explain clearly and as simply as possible, but sticking to established expository forms to avoid the work’s being rejected due to too much novelty.

RHollerith 1 Feb 2026 20:59 UTC
2 points
0
in reply to: Oliver Daniels’s comment on: An Explication of Alignment Optimism
Increasing the volume of alignment research helps only if alignment research does not help capabilities researchers, but my impression is that most alignment research that has been done so far has helped capabilities researchers approximately as much as it has helped alignment researchers. Just because a line of research is described as “alignmnent research” does not automatically cause it to help alignment researchers more than capability researchers.

In summary, I don’t consider your (a) a cause for hope because the main problem is that increasing capabilities to the point of disaster (extinction or such) is easier than solving the alignment problem and your (a) does not ameliorate the main problem much if at all for the reason I just explained.

RHollerith 31 Jan 2026 22:32 UTC
5 points
1
on: An Explication of Alignment Optimism
Suppose AI assistants similar to Claude transform the economy. Now what? How is the risk of human extinction reduced?

Yes, alignment researchers have become more capable, and yes, the people trying to effect an end or a long pause of AI “progress” have become more capable, but so have those trying to effect more AI “progress”.

Also, rapid economic change entails rapid changes in most human interpersonal relationships, which is hard on people and according to some is the main underlying cause of addictions. Addicts aren’t likely to come to understand that AI “progress” is very dangerous even if they are empowered by AI assistants.

Your argument supports the assertion that if AI “progress” stopped now then we’d be better off then we would’ve been without any AI progress (in the past), but of course that is very different than being optimistic about the outcome of AI tech’s continuing to “progress”.