davidshepherd7

Did you try Opus-4.7 on a lower reasoning level? Looks like on `max` it's using far more tokens than the other frontier models.

That page is copied verbatim from https://unsung.aresluna.org/molly-guard-in-reverse/ (which is linked at the top). The original page also has much better formatting.

On the off chance that the author reads this: can you enable an RSS feed please?

I want to subscribe, but I never end up reading newsletters if they land in my email inbox.

A mirror that worked for me: https://skyview.social/?url=https://bsky.app/profile/robpike...

This is a perfect replacement for me, thanks for recommending it.

IIUC chainguard is this, but only for python, java, and docker images so far. https://www.chainguard.dev/libraries

Could you give some examples of more secure solutions?

In case it's useful to anyone: another implementation of this idea is Weeve https://shop.weeve.ie I bought one of their books (a study in scarlet) but it wasn't great. Lots of mistranslation, especially later on in the book. The general idea seemed to work well though, with better implementation I think it could really help my french.

I'll give this one a try, being able to add my own books is particularly exciting.

Does anyone have a list of which products are in which business units?

Yeah I think this is correct with the edit.

I guess this is mostly relevant for software that runs on shared infra, sends requests to a url provided by an attacker (e.g. webhooks), and uses a SOCKS5 proxy?

The article was too vague for me to really understand, but I think it's suggesting something like Brandur's deeply-integrated idempotency keys design [1] except that the infra is provided by a saas company? Also somehow it supports rollbacks?

Does anyone know of any other more concrete explanation of how these kinds of systems could work?

[1] https://brandur.org/idempotency-keys

This sounds really interesting! Can you explain or link to an explanation of the technical defenses that Google has?

(Or, is this comment https://news.ycombinator.com/item?id=35585453 complete and accurate?)

They actually use a python monolith unless it's changed recently. See e.g. https://instagram-engineering.com/static-analysis-at-scale-a...

The company I work for has saved people and businesses in West Africa roughly $1B in banking fees this year.

It's a big company but I'm really proud to have done my part towards this.

Any use of OS features by calling functions in the standard library should work though, right?

I think it's only dependencies that contain C code that would need further changes. I think you would have to recompile then with cosmopolitan libc.

I'm curious how this relates to givedirectly's research https://www.givedirectly.org/research-at-give-directly/

As far as I know givedirectly have found that cash transfers are effective at improving people's long term wealth. But I've never heard anything about them moving people from the poor to the "less poor" mode of a bimodal wealth distribution. Maybe the difference is that this experiment was run in a place that has a clear bimodal distribution, which allowed them to observe this more easily?

Wish I could read the paper but it's behind a paywall and not on scihub.

I'm probably missing something, but it sounds like using Warp has a bunch of downsides vs "just" creating a read only replica using logical replication and then failing over. Did you choose Warp only because of Azure's limitations or were there other reasons?

Is it just me or are all the diagram links broken? It's pretty hard to understand without them.

Do you mean using spaced repetition to memorize checklists or in general?

Another problem with the current situation which isn't mentioned in the PEP: I often see people writing `foo_bar or 1` instead of `foo_bar if foo_bar is not None else 1`, which is usually incorrect when a is a number (zero is falsey). I think a terser "official" way of doing this would help there.

But, I often solve this with a function like:

``` def default_to(x: Optional[T], d: T) -> T: return x if x is not None else d ```

I wonder if having this function as the official way to do it would solve the problem without the "specialness" of `??`.

But there's no good equivalent that I know of for `?.`, that's a real pain to do with python as it is now so it seems like a nice addition.

Are you sure? The graph in the article seems to match a graph on the ARA's website: https://www.ara.fi/en-US/Materials/Homelessness_reports/Repo... (the ARA seem to be the organisation in charge of housing in Finland).

Not that it's completely reliable, but Wikipedia also has information which matches the article (https://en.wikipedia.org/wiki/Homelessness_in_Finland), no discussion on the talk page, and no signs of an edit war in the history.

> It's exhausting, there's basically no downtime, and we end up in spirals of shame and guilt at being terrible parents, partners, employees. .... our kids are sat in pyjamas on the Nintendo Switch at 2pm ...

I've been reading "Selfish reasons to have more kids" (Brian Caplan) recently in semi-preparation for becoming a parent, and, well I'll just quote the book summary

> We've needlessly turned parenting into an unpleasant chore. Parents invest more time and money in their kids than ever, but the shocking lesson of twin and adoption research is that upbringing is much less important than genetics in the long run. These revelations have surprising implications for how we parent and how we spend time with our kids. The big lesson: Mold your kids less and enjoy your life more. Your kids will still turn out fine.

I've been using Anki for a few years, here are my thoughts on your questions:

> I have never really found that memorization was ever an obstacle to my ability to learn something.

I think it depends what you are learning, for example when learning languages I find a lot of the difficulty is just in retaining the words.

Another place where I've found it useful is for memorizing definitions in (pure) mathematics. To be able to read the later parts of a book you need a good working knowledge of earlier definitions. You could look back for the definitions every time, but that breaks the flow of reading.

> I've always found that retention proceeds from understanding.

Yeah, I find that too. But this is something that I do after understanding something to help retain it better and for longer.

> Time spent flicking through flashcards for vocabulary or formulae or chemical paths or whatever seems like a poor substitute for time spent reading material that uses the vocabulary ...

The time I spend on memorization is very small, less than 5 minutes a day, and I usually do it on my phone while I'm waiting for something. It replaces checking reddit or playing phone games rather than deep reading of interesting material.

Just to back this up: the journal it's published in has a very very low impact factor of 0.235 (i.e. it publishes 4x more papers that it gets citations, probably most of it's papers are never cited by anyone) [1]. If this paper was considered a big a deal by the wider research community you would expect it to be published in a high profile journal.

[1]: https://journal.komci.org/ViewJournalInfo.php?JID=109#Impact...