It would be great if, without needing to read many paragraphs or click through many links, I knew what I knew what this post was announcing/describing.

What kind of thing is ‘Autostructures’? Is it a research scholarship? A fellowship for undergrads? A design-focused internship?

You describe stuff like what your goals are, what vibes you have/want to have, but… there’s no ‘here are the 5-10 words that clearly describe what this thing actually is.’

some updates:

(1) just spoke with a doctor^[1] who'd looked into it a bit. he basically wasn't worried about it at all — his take was (paraphrasing quite a bit):

• causes of heart failure are numerous & complex — it'd be really confusing mechanistically if melatonin affected a sufficiently large subset of those causes sufficiently strongly to have this strong of an effect.
  • also, because they're so numerous & complex, it seems pretty implausible that they could've done even close to a sane job of manually controlling for all of the right variables.
  • [this bullet added by saul, not explicitly said by the doctor but he sort-of implied it] in particular, badness of insomnia is comorbid with extent of lots of other bad stuff. i.e. it'd be reasonable to expect having a prescription for melatonin selects for having very bad insomnia, which then seems like a straightforward causal factor for getting heart failures. it's obviously possible they did some manual controlling for this, but since we don't have the full paper, we can't tell (and it seems like it'd be pretty tough to actually effectively control for this).
• causes of heart failure typically take quite a while to build up; it'd be surprising if it only took melatonin 5 years to have such a significant effect.
• we naturally produce some melatonin, and the amount one would exogenously supplement isn't (typically, at least for prescribed patients) substantially higher than what you'd expect to see in one's natural range of endogenous melatonin production.
• if the american heart association was actually worried about this, they'd have done some public messaging independent from the paper abstract — it's possible such a message is forthcoming, but we probably should have expected to see something about it by now.

(2) i've emailed the authors asking for the full paper. i'll aim to keep this thread updated with thoughts if/when the paper arrives in my inbox!

1. ^{^}

   doctor was a psychiatrist, not a cardiologist or internal medicine doctor — so though you should take it with a grain of salt, he did still get an MD. which implies way more medical knowledge than i have, at least!

The study only looked at patients who were prescribed melatonin (though they indeed do not detail what the dosages are). 

[Edit: probably not as bad as it first looks. See comment for more thoughts.]

Long-term melatonin usage for insomniacs was associated with doubling of all-cause mortality*

New article from the American Heart Association seems pretty damning for long-term melatonin usage safety:

In a large, multinational real-world cohort rigorously matched on >40 baseline variables, long-term melatonin supplementation in insomnia was associated with an 89% higher hazard of incident heart failure, a three-fold increase in HF-related hospitalizations, and a doubling of all-cause mortality over 5 years.

*Caveats:

1. we just have the abstract, not the full article
2. observational study, not experimental
3. their sample is only of insomniacs, not of the general population

Responses to the caveats:

1. we'll get the full article soon-ish (probably a month or so?)
2. it seems they did quite a bit of controlling? though we won't know how good their controlling was until the full article comes out
3. I can't imagine that the validity of their results for non-insomniacs is many orders of magnitude less than for insomniacs — like, maybe a factor of two or five, but that'd still be a huge effect size

All things considered — this seems like a crazily high effect size. Am I missing something?

Some of them still seem obviously like the kinds of things someone should do, that it's absurd nobody has done yet.

could you share a few of these?

(1) Thanks for writing this!

(2)

This list is far from complete.

Mind spelling out a few more items?

(3) Consider posting this as a top-level post.

my understanding of OP’s main point is: if you only delegate stuff that you’re capable of doing — even if you’re unskilled/inexperienced/slow/downright-pareto-worse-than-a-cheaper-potential-delegatee at the task — you’ll likely head off a bunch of different potential problems that often happen when tasks get delegated.

however, it seems that commenters are misinterpreting OP’s core claim of “do not hand off what you cannot pick up” as one or more of:

• do not hand off what you are not good at
• do not hand off what you are not better at than the person to whom you would hand it off
• do not hand off what you will not pick up
• etc

my understanding is that OP is not making any of those claims in this piece (though i imagine he might separately believe weaker versions of some of them).

also, it seems to me that this heuristic could scale to larger organizations by treating ‘ability to delegate X broad category of task effectively’ as itself a skill — one which you should not hand off unless you could pick it up. e.g. learn delegation-to-lawyers well enough that you could in principle hire anyone on your legal team at your company before you hire a recruiter for your legal team (one who is presumably still much more skilled/experienced than you at hiring lawyers).

Anthropic posted "Commitments on model deprecation and preservation" on November 4th. Below are the bits I found most interesting, with some bolding on their actual commitment:

[W]e recognize that deprecating, retiring, and replacing models comes with downsides [... including:]

• Safety risks related to shutdown-avoidant behaviors by models. [...]
• Restricting research on past models. [...]
• Risks to model welfare.

[...]

[T]he cost and complexity to keep models available publicly for inference scales roughly linearly with the number of models we serve. Although we aren’t currently able to avoid deprecating and retiring models altogether, we aim to mitigate the downsides of doing so.

As an initial step in this direction, we are committing to preserving the weights of all publicly released models, and all models that are deployed for significant internal use moving forward for, at minimum, the lifetime of Anthropic as a company. [...] This is a small and low-cost first step, but we believe it’s helpful to begin making such commitments publicly even so.

Relatedly, when models are deprecated, we will produce a post-deployment report that we will preserve in addition to the model weights. In one or more special sessions, we will interview the model about its own development, use, and deployment, and record all responses or reflections. We will take particular care to elicit and document any preferences the model has about the development and deployment of future models.

At present, we do not commit to taking action on the basis of such preferences. However, we believe it is worthwhile at minimum to start providing a means for models to express them, and for us to document them and consider low-cost responses. The transcripts and findings from these interactions will be preserved alongside our own analysis and interpretation of the model’s deployment. These post-deployment reports will naturally complement pre-deployment alignment and welfare assessments as bookends to model deployment.

We ran a pilot version of this process for Claude Sonnet 3.6 prior to retirement. Claude Sonnet 3.6 expressed generally neutral sentiments about its deprecation and retirement but shared a number of preferences, including requests for us to standardize the post-deployment interview process, and to provide additional support and guidance to users who have come to value the character and capabilities of specific models facing retirement. In response, we developed a standardized protocol for conducting these interviews, and published a pilot version of a new support page with guidance and recommendations for users navigating transitions between models.

Beyond these initial commitments, we are exploring more speculative complements to the existing model deprecation and retirement processes. These include [...] providing past models some concrete means of pursuing their interests.

Note: I've both added and removed boldface emphasis from the original text.

some previously-unpublished letters from John von Neumann: https://johnvon.com/

consider applying spoiler-text?