SubiculumCode

I am no expert, but I think it is this: 1) We have few effective tools at monitoring alignment right now, and chain of thought is one of the more effective. 2) Monitoring latent space may be possible, but I do not think it is even close to being a solved problem, nor whether it is possible at scale and outside of controlled problem areas. 3) Finally, more recursion within latent space may complexify the latent representations, not simplify them.

It is still helpful I believe, but your point is well taken. The problem is that we have relatively few tools for monitoring alignment, and longer loops of processing that stay in latent space means less ability to monitor.

The major concern with looped transformers is that makes it more difficult to monitor model alignment. When more processing occurs within latent space without outputting text, that means less effective, frequent chain-of-thought monitoring, and the potential for greater un-monitored latent-space shenanigan.

Alpha Fold has continued to impact the field of protein networks, but I do hear that not every one of the deep learning biology models from Google/Deep Mind and others have made equivalent impact or had as lasting relevance in their respective domains..some have performed more poorly than other available models. I'd love to learn more about this, but this has mostly come from little snips of conversations here and there, in person and online, but I haven't seen anything comprehensive in terms of evaluating their impacts overall

Does math have an end? Or is it an infinite self-construction?

So California wet year. We don't mind a good snow pack, but I've become increasingly worried about catastrophic flooding evens in Cali e.g. [1], and this does not ease those worries.

[1] https://www.science.org/doi/full/10.1126/sciadv.abq0995

let's give drivers motion sickness isnt a good strategy, just my opinion.

There is interesting evidence that autism-autism social interactions are fairly vibrant. Everything is a bit more complicated than simple answers, no matter which way you swing.

I agree 1000% that the full spectrum of autism must be included into autism research.

My research for the last 10 years has been dedicated to collecting and analyzing data for the Autism Phenome Project [1]. Unlike almost all autism neuroimaging research, we include the full spectrum of autism, with a third of our participants with severe intellectual disabilities, closely matching their share of the autistic population, as well as those who are non-verbal, have extreme specific phobias and anxiety, sensory sensitivities, self-injurious behaviors: the full spectrum. It still frustrates me to no end that almost every neuroimaging study you read by default excludes those with IQs less than 80, while our cohort has managed, through very hard work, managed to image individuals with IQs in the 20-60 range with regular success.

[1] https://pmc.ncbi.nlm.nih.gov/articles/PMC8801875/

First, and foremost, the rise in autism rates is due to more sensitive instruments, a broadening of some inclusion criteria between DSM 4 and 5, more awareness of autism, less stigma associated with autism, services attached to autism diagnosis. I believe current research suggests these don't account for all the rise, but does account for most of it.

If I were to include additional biological possibilities, I'd include: a) birth control -> waiting to have children until a later age, more de novo mutations due to age. b) assortive mating, since some careers tend to attract those on the within and on the edges of the spectrum., c) perhaps controversially, and with the very little research on the subject, endocrine disruption from plastic waste exposures.

Genetic contributors to autism include:

1. common variants. Common variants are genetic variants weakly associated with autism that we all have to some degree or another, and that only result in autism given a sufficient number of them are found in a person..or at least that's a common idea and pretty well supported. These variants are inherited.

2. rare de novo variants: These are spontaneous genetic variants that the parents did not have, but there was a mutation during production of sperm, for example. Many times, rare de novo variants that occur in highly conserved genes cause tremendous neurodevelopmental problems, spanning death often prenatally, other conditions, and for, autism, most frequently with severe intellectual disability.

Common variants are heritable and common, and are part of the normal spectrum. Only with a very large load do these result in the severest conditions, I think, like intellectual disability.

Occurrence of rare de novo variants are associated with things like age of father (older, less reliable production of sperm without genetic mutations), pesticide exposures, etc.

There is also a real immune factor involved, as well as environmental exposures, which can interact with genes.

A lot of the behaviors by which autism is diagnosed (social communication, repetitive behaviors and restricted interests) are fairly high-level behaviors, dependent on a lot of cognitive/neural machinery beneath it. There is an idea out there that convergent behaviors that allow consistent diagnosis depends on this, as the resulting behaviors might results from differences in many different underlying processes, but the result at the higher level can be measured the same.

Analogy: Two programs can produce nearly identical behavior at the UI level, but be implemented very differently. Diagnosis operates partly at the "interface" level, but what happens underneath can differ greatly.

no sarcasm tags needed unfortunately, and we autism scientists are very worried/ For example, as I submitted a few days ago: https://news.ycombinator.com/item?id=49422954

Autism is a spectrum. Neurodiversity within and outside of autism is both intrinsically valuable and extrinsically valuable to society. However, not all phenotypes of autism should be viewed as neurodiversity, but as harmful. There is only so much that society can do to make severe intellectual impairments less disabling without identifying a cure for the intellectual impairment. The debate between medical versus disability models, neurodiversity versus pathology, inevitably gets pushed by those that want to classify everything into clean essentialist boxes, but that, I do not believe, is a helpful way to view the world. Not every condition is a pathology, nor is every condition benign if only society would do more to make it less disabling.

autism researcher here. I have not yet read this study. It will not be the kind of paper that many, including myself, can read quickly and understand. Autism is a multi-disciplined science, so there is a lot to understand. However, my initial sense from reading the abstract is that this paper might be a really really big deal and incredible new resource.

General problem statement: Many different genetic variants are associated with autism, some strongly, most very weakly. Many have suspected that this plethora of genetic variants might converge onto a smaller set of biological pathways affecting neurodevelopment in ways that results in an autism phenotype or phenotypes. This has been our hope, because the alternative scares the hell out of a scientist. The saying goes that if you met one person with autism, you've met one person with autism...meaning there is a ton of heterogeneity within the autism spectrum, even if a certain set of core autism symptoms are displayed. So there was a existential crisis that there is no "autism", just a billion autisms, where your sample size is always N=1. By demonstrating evidence that many variants converge on the same biological pathways, there is hope that understanding autism is tractable.

Yeah, I suppose it wasn't specifically relevant to Yayoi, but I was only giving an example of why my hypothesis might be true, not to restrict it to computer-related persons.

I once thought I'd test the hypothesis that the frequency of HN front page stories reporting a death was increasing over time. Started figuring out how to get the data, got distracted by life, and never got back to it.

edit: that was not a criticism of such stories, but I was alluding to the fact that the generation that pioneered early computing in the 70's and 80's are getting older.

just a FYI, I am not behind this submission or involved in anyway. I just thought it was curious, and wondered about it.

Did you all notice this submission? Super Curry Combat II [1][2]

[1]https://news.ycombinator.com/item?id=49454496

[2]https://curry-combat-2.com/

Fast as AI turnaround indeed?

Is this a demonstration of a fast-as-ai turnaround?

How did you do that? I noticed the spacing was very wide between leters,it bothered me.

I have two bidets, neither use electricity. Also they fit on existing toilets. Just search bidet kit.

Once you go bidet, you don't go back. There are lots of nice kits that you can install on your existing toilet too, making it super easy. [1] is a generic Amazon search. Electricity not required for most

We use so much less tp since we got one.

[1] https://www.amazon.com/s?k=bidet+kit

Playing Chess without letting it write a chess engine.

I can't wait for an actual competitor to the M chips from apple. It's frustrating.

Can we please go back to letting experts lead departments where expertise is paramount, you know, like leading scientific funding?

Am I reading it right that they are claiming about ~%15 word prediction accuracy?

A conspiracy theorist misuse a term with no knowledge of it's context? Nahhh never /s

I wonder if these are what was being referenced: [1]https://pmc.ncbi.nlm.nih.gov/articles/PMC11092452/ [2]https://www.nature.com/articles/s41592-025-02932-8

ChatGPT helped me identify likely articles: https://chatgpt.com/s/t_6a8a7e1c82308191be82762907a607de

Not an expert, but isn't your LCDM vs MOND example assymetric? Like different kinds of data supports each model very well, and so reconciliation isn't simple?

Moreover, your objection that science progresses one funeral at a time is exaggerated, even if it is real, it's not necessarily all that strong: https://pmc.ncbi.nlm.nih.gov/articles/PMC6814193/