Morpheus

Karma: 972

Morpheus Aug 15, 2025, 12:17 AM
24 points
13
on: Morpheus’s Shortform
I am in the waiting room at the doctor and the ~1 year old child next to me is scrolling through YouTube shorts on his mother’s phone with the mom watching along. Incredibly incoherent AI slop. Not an expert at early development, but this seems very suboptimal. I know that scrolling is not good for my reward system, but a 1-year old? At least put some random video on instead of letting the child scroll? If I was in charge at Google and was living by “don’t be evil”, I would maybe make a classifier to identify children scrolling like this and giving a reminder once in a while to suggest some alternative activities to the parent?

Morpheus Aug 10, 2025, 2:29 AM
1 point
0
on: Mechanistic Interpretability is Being Pursued for the Wrong Reasons
Thanks for writing this post! I do think some people are pursuing interp for this “wrong reason” of trying to prevent scheming, and the road where you get interp to improve that well to make that work seems unlikely (understanding general circuits doesn’t seem impossible to me but extremely hard and nonzero people are working on this).

I think that perhaps the mistake comes from mistaking the simplicity of the optimizer for a property of the mesa-optimizer. SGD by backprop is one algorithm so people put a single label, “deep learning,” on all models it produces. But there is no reason that all of these models must use similar circuits. They may all use an array of unique fantastically complex circuits. Understanding every circuit that can be produced by SGD at once is not a cohesive research program, and it is not a plan that will succeed.

There would be reasons to believe that models are going to use similar algorithms if they use similar training data. Understanding every circuit that is possible to be produced by SGD given “infinite training data” seems intractable, but “in practice” I’d expect different algorithms produced by SGD to produce modular structures with common “motifs” just like evolution does. Evolutionary developmental biology is indeed a field (that just like interp seems more bottlenecked on better theory rather than measurement capabilities). It’s why I am still excited about developmental interpretability, even though I don’t have a coherent plan for how it will help us with safety beyond “more theory and foundations” seems nice (with the general caveat that I am very confused about capability externalities, but this seems kind of unavoidable for actually broad insights).

Morpheus Aug 1, 2025, 4:38 AM
20 points
11
on: Do Not Render Your Counterfactuals
The fact that this type of thing tends to get such large emotional responses out of people makes me wonder to what extent rendering counterfactuals would be useful to combat the lack of imagination for big decisions that actually are ahead of someone?

Morpheus Jul 31, 2025, 7:34 AM
2 points
0
in reply to: AnnaJo’s comment on: AnnaJo’s Shortform
I would get a vaccine again if I thought I was at the risk of getting it?

Morpheus Jul 30, 2025, 11:56 PM
1 point
0
in reply to: Dacyn’s comment on: Sex Determination as a Bottleneck to Species Development
Yes! For the individual, it does not make sense to adjust the sex ratio with changes in climate, but for the species overall it’s not that bad or even positive. Sex chromosomes do a much better job for the individual here (And would be selected for if temperature changes were happening too often). I do think I had confused thinking at the time, because I also had just read about a theory in humans that high status females supposedly produce more males (so payoff could be different between low and high status, but that doesn’t apply in this case, and it’s not born out well empirically anyway).

What I do think is true: It might not be that much of a coincidence that climate change would lead to more females. If it was the other way around, there is a higher likelihood the species would have gone extinct from a meteor, and we wouldn’t observe it today. Having 1% females and 99% males is going to lead to a really bad population bottleneck, but having 1% males and 99% females just means the remaining males have little competition (the effective population is still reduced to some extent). The fact that turtles don’t fit this pattern and look like they should have gone extinct with the last meteor tells you how weak and slow selection between species is.

I was thinking of this when I saw these articles complaining how sex ratios are supposedly endangering sea turtles. It doesn’t seem obvious to me that the sex ratios alone should be a problem (although ¹⁄₉₉ is probably bad).

Morpheus Jul 30, 2025, 10:23 PM
1 point
0
in reply to: leerylizard’s comment on: Sex Determination as a Bottleneck to Species Development
I thought about this a bunch of years ago, and I am not quite sure how much this matters. Often ants get sperm from multiple fathers, and then they’d be less related than siblings (although maybe they only start doing that once they are out of the unstable intermediate period between being not eusocial and being eusocial). Also, this leads to colonies producing fewer males than would be game theoretically optimal for the queen and to conflict between queen and workers (50/50 would be optimal), although that might be helpful on the species level? So I think most of the cohesion benefits from 75% kinship between daughters is offset by the 25% kinship between brothers and sisters and if the queen mates with 2 males instead of 1, then the kinship is actually less than with monogamous termites (which is most termites, while I think most ant queens mate with multiple males).

I do think the difference in kinship could make it easier to develop eusociality, but I think my above argument that you just need a queen, so adapting behaviour is a simpler program might explain better how ants got through the unstable period between being individualistic and being sort of eusocial better (also if you start out with two parents there is less marginal returns to another worker for the nest, so that might also explain why it happened so often in Hymenoptera rather than other nest building insects). There are some wasp species that sometimes decide to build nests together and do division of labour (protect nest, search food) even though they are unrelated, but I forgot their name (Most of my opinion above is informed by the Superorganism book from EO Wilson. He has some idiosyncratic beliefs when it comes to group selection, but I think when it comes to his analysis that the kinship seems marginally important, he seems right). The kinship could explain why ants developed it multiple times, but it doesn’t explain why termites even after developing eusociality didn’t expand further.

Morpheus Jul 30, 2025, 3:55 PM
3 points
0
in reply to: Thane Ruthenis’s comment on: johnswentworth’s Shortform
I tend to have some luck with concrete analogies sometimes. For example I asked for the equivalent of Tonedeff (His polymer album is my favorite album) in other genres and it recommended me Venetian Snares. I then listened to some of his songs and it seemed like the kind of experimental stuff where I might find something I find interesting. Venetian Snares has 80k monthly listeners while Tonedeff has 14K, so there might be some weighting towards popularity, but that seems mild.

Morpheus Jul 27, 2025, 9:56 PM
3 points
2
in reply to: Linda Linsefors’s comment on: Linda Linsefors’s Shortform
I think when you wrote postmortem you meant to write premortem?

Morpheus Jul 24, 2025, 1:22 AM
1 point
0
in reply to: tailcalled’s comment on: leogao’s Shortform
Sounds interesting and like something I might miss if true. I would be interested in examples.

Morpheus Jul 23, 2025, 10:02 AM
1 point
0
in reply to: StartAtTheEnd’s comment on: Generalized Hangriness: A Standard Rationalist Stance Toward Emotions
I was engaging with this, because I thought maybe you are advocating for some kind of doubles think, and I might spell you out of that, but this doesn’t seem to be the case. I am not interested to get deep into that religion argument (too many different people and different religions). Yes, there are some topics like ethics where most people don’t benefit from reasoning about them explicitly, and even smart people tend to get very confused by them. I remember I was confused for days by my first introduction to ethics.

Morpheus Jul 21, 2025, 7:48 PM
1 point
1
in reply to: StartAtTheEnd’s comment on: Generalized Hangriness: A Standard Rationalist Stance Toward Emotions
Reframing your thoughts such that you don’t step on your own toes is great. I am not quite sure what you are trying to argue for with the religion point. Do you actually endorse this on a practical basis? Have you convinced yourself that you need to yourself belief in untrue things or that it is good if other people are believing untrue things in just the right way they cancel each other out? Believing in God seems a particularly bad example. Having high alieve you can solve a problem that lots of people haven’t solved before you might be fine for minutes or a few days. But I can’t see how you would get psychological benefits from believing in a god or fate and whatever and that not messing up your epistemics.

Morpheus Jul 18, 2025, 9:43 AM
3 points
0
on: Foom & Doom 2: Technical alignment is hard
Thank you for this post! My impression is that this post makes real progress at identifying some of the upstream cruxes between people’s models of how AGI is going to go.
As examples, if you look at @So8res’s AGI ruin scenarios are likely (and disjunctive), I claim that a bunch of his AGI ruin scenarios rely on his belief that alignment is hard. I think that belief is correct! But still, it makes his argument less disjunctive than it might seem. Likewise, I now recognize that my own What does it take to defend the world against out-of-control AGIs? sneaks in a background assumption that alignment is hard (or alignment tax is high) in various places.
This seems correct to me.
But I figured out that I can occupy that viewpoint better if I say to myself: “Claude seems nice, by and large, leaving aside some weirdness like jailbreaks. Now imagine that Claude keeps getting smarter, and that the weirdness gets solved, and bam, that’s AGI. Imagine that we can easily make a super-Claude that cares about your long-term best interest above all else, by simply putting ‘act in my long-term best interest’ in the system prompt or whatever.” Now, I don’t believe that, for all the reasons above, but when I put on those glasses I feel like a whole bunch of the LLM-focused AGI discourse—e.g. writing by Paul Christiano, OpenPhil people, Redwood people, etc.—starts making more sense to me.
That seems to represent well the viewpoint that I now have even less faith in, but that didn’t seem strictly ruled out to me 4 years ago when GPT-3 had been around for a while, and it seemed likely we were headed for more scaling:
39 Nothing we can do with a safe-by-default AI like GPT-3 would be powerful enough to save the world (to ‘commit a pivotal act’), although it might be fun. In order to use an AI to save the world it needs to be powerful enough that you need to trust its alignment, which doesn’t solve your problem.
- What exactly makes people sure that something like GPT would be safe/unsafe?
- If what is needed is some form of insight/break through: Some smarter version of GPT-3 seems really useful? The idea that GPT-3 produces better poetry than me while GPT-5 could help to come up with better alignment ideas, doesn’t strongly conflict with my current view of the world?

Morpheus Jul 12, 2025, 3:29 PM
5 points
1
in reply to: Hastings’s comment on: Raemon’s Shortform Feed
Emacs has Eliza still built in by default of course :)

Morpheus Jul 12, 2025, 6:17 AM
3 points
0
in reply to: Ruby’s comment on: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity
1. This is much less true of my participation in the study where I was more conscientious, but I feel like historically a lot of my AI speed-up gains were eaten by the fact that while a prompt was running, I’d look at something else (FB, X, etc) and continue to do so for much longer than it took the prompt to run.
The temptation to multitask strikes me as a very likely cause of the loss of productivity. It is why I virtually never use reasoning models except for deep research.

Morpheus Jul 12, 2025, 5:33 AM
2 points
0
in reply to: Isopropylpod’s comment on: Isopropylpod’s Shortform

And on a more micro-level, living knowing that I and everyone else have one year left to live, and that it’s my fault, sounds utterly agonizing.

Earlier you say:

or frankly even if anyone who continues to exist after I die has fun or not or dies or not, because I will be dead, and at that point, from my prospective, the universe may as well not exist anymore.

How are these compatible? You don’t care if all other humans die after you die unless you are responsible?

Morpheus Jul 6, 2025, 10:30 PM
3 points
0
in reply to: johnswentworth’s comment on: Core Pathways of Aging
Why wait for mice? If I understand correctly, a lot of the symptoms common to aging also show up in Saccharomyces cerevisiae. The goal of the synthetic yeast project 2.0 was among other things creating chromosomes without retrotransposons, and they recently finished synthesising all 16 chromosomes, so all that is left should be assembling them? If those synthetic cells do or don’t show typical signs of aging, that should be a strong sign either way, right?

Morpheus Jul 4, 2025, 12:00 PM
14 points
2
in reply to: Kabir Kumar’s comment on: Kabir Kumar’s Shortform
Deep Research found this PDF. Search for “I was wrong” in the PDF.

Morpheus Jun 26, 2025, 12:16 PM
5 points
0
in reply to: johnswentworth’s comment on: johnswentworth’s Shortform
The last year, my median conversation was about as entertaining as yours. The top 10% conversations are fun-in-their-own-right at that moment already because my brain anticipates some form of long-term value (with the exception of cracking jokes). I don’t know if all those conversations would count as “casual”. As intellectually stimulating as the Task Master TV-show is funny. Conversation is more heavy tailed than movies though. Long term value includes: learning or teaching (learning some new technical thing that’s usually not written down anywhere (Podcasts tend to be better for that), getting a pointer about something to learn about, teaching something technical in the anticipation that the other person is actually going to do anything with that knowledge, incorporating the generating function behind someone’s virtues/wisdom), thinking out loud with someone else in the expectation that this might lead to an interesting idea, gossip, life stories (sometimes preventing you from harm from people/situations that can’t be trusted. Sometimes just illuminating parts of life you’d know less about). My most fun conversation had me grinning for 30 minutes after still, and my heartbeat after that time was also still 10–20 beats higher than usual.

My median conversations at parties over my entire life are probably less entertaining than your median ones. My bar for an interesting conversation also rose when I stumbled over the wider rationalist sphere. I remember two conversations from before that era where the main important information was essentially just “there are other smart people out there, and you can have interesting conversations with them where you can ask the questions you have etc.”. One was at a networking event for startup founders, and the other was a Computer Science PhD student showing me his work and the university campus (same conversation that got my heart-beat up).

Morpheus Jun 18, 2025, 9:08 PM
4 points
0
in reply to: Roman Malov’s comment on: Morpheus’s Shortform
I was just trying to adjust a loophole that often seemed to be missing in Umeshisms, but I think this made my statement more confusing: if you are 15 years old (the particular age is irrelevant, I am just saying an age exists), then you having sent 1-2 cold emails is not too little, nor did you invest too much time, you are just young and there weren’t that many worthy occasions yet. If you have just taken a single flight in your life and missed 0, this is not large evidence that you spend too much time at airports.

Morpheus Jun 18, 2025, 3:22 AM
1 point
−2
on: Morpheus’s Shortform
If every private message on lesswrong or every cold email you ever wrote has received a response, you are either spending too much time writing them, are very young or aren’t sending enough of them.