calpaterson

Interestingly, after writing this (some years ago) I spoke to some of the original authors. They had never used Smalltalk. So I suppose they invented this stuff independently

No, no, there is a single world ring. the 16mb limit is for values

Well, I say "more or less" :). But the fact that there is a single global database doesn't mean that you can read every key or value (but generally, yes, you can _read_ everything). I mention in passing prolog-style permission systems for evaluating perms.

But anyway, specific trades are rarely private to one part of the bank for many reasons. For example regulatory: these days you have to notify the regulator about every trade.

They don't use pip, you just import the module and it is pulled from barbara

Mostly is no rule, adding a cache can just save you from having to buy a bigger database instance in many cases.

The most common first thing to cache is getting the current user, because this ends up being a very hot path for most stateless systems. Because you need to get the current user for almost every request, it's quite easy for getting the current user to be 50% of database load: first you get the user, then you do the thing. tada, user lookup is now half your app by volume

Once someone decides they want to use redis as something other than a cache, you sort of do have 2 cache technologies anyway. You can't use a redis instance that is configured for caching for any other purpose (caching instance must have eviction, non-caching instance must not have eviction). You need a second redis with a different configuration.

Honestly designing your app to have a "memcache-friedly cache layout" is the same thing as designing it to have a redis-friendly cache layout. The pattern for this kind of application cache is identical: "get, and if not there, calculate and set".

Yes I switched from claude code to opencode with deepseek recently.

It is basically indistinguishable from sonnet. At this point my own prompts, AGENTS.md, background docs and so on matter a great deal more than the differences between models.

And deepseek v4 flash (the sonnet comparable) costs 3% of what sonnet does.

It often is illegal to pay them. They are often on sanctions lists, or indeed in embargoed countries. And it's just generally not allowed to pay unidentifiable parties for basic anti-money laundering reasons. And a lot of countries are bringing in new legislation to make paying illegal, starting with public sector organisations. I'm sure that will only expand.

Frankly, you pay a ransom at your peril. If it turns out it was North Korea you may well go to jail for it.

Fair enough. I didn't say so but - bookstack is great. Thanks so much for it.

As best I can see, bookstack has not experienced much in the way of concrete issues with Github. And there there are no concrete benefits from migrating to Codeberg. It is his project, his perogative, completely. But the major disbenefit of being on some other forge is that people are less likely to find the software and so less likely to adopt it.

I use Bookstack for a family wiki. I probably would not have gone with it if it had not be hosted on Github as the visible activity on Github makes it clear that it's a project with momentum (18k stars, lot's of activity) etc.

I can't help but feel that moving will make the project less successful than otherwise...

"56k" meant 7 kilobytes per second as a theoretical max. So 4.4 was ok. Everything with networks is done as bits, I think honestly for marketing reasons now

Interesting process. I wonder if he considered doing this with Anki. That would have given him a good SRS algo for free and Anki cards are also HTML+CSS+JS. I probably wouldn't try to put LLM calls onto my cards though

The only oversight I think in the proposal is staggered distributions so that projects declare a UUID and the distribution queue progressively makes it available rather than all or nothing

That is indeed an oversight - I wish I had thought of that idea!

Fair, but it is part of a pretty large library system and you can order whatever you want to pickup at Oodi

I have heard of more than a few horror stories including filesystems lost and force pushes done.

These tools have only been in use for a short time and the current harnesses/system prompts are quite limited. Claude code is mostly limited to your codebase where you have version control. Excel is different.

I foresee that once people hand over more power to full agents there will be some nasty surprises. Im sure there will eventually be demand for some kind of limits

And yet I could tell them apart with pretty good confidence. Why?

I got 8/9. Been away 2 years. The ones I rarely used or which don't have obvious "tells" that are hardest. For me, Jubilee is the most obvious - very distinctive sound.

Yes, I of course link to this post, which I think is great. But I think actually it understates the case. All three parts of the trifecta (untrusted content, private data and external comms) are not necessary. Really, the key problem is just untrusted content in the context window. Access to private data and the ability to communicate externally are just modalities in which damage can occur.

For example: imagine having just untrusted content and private data (2/3 parts of the trifecta). The untrusted content can use a "Disregard that!" attack to cause the LLM to falsely modify the private data. So I think the whole "trifecta" is not necessary and the key thing is that you simply can't have untrusted stuff in your context window at any point.

I agree and one of the things that makes it harder to handle "disregard that!" is that many models for LLM deployment involve positioning the agent centrally and giving it admin superpowers.

I mention in the footnotes that I think that it makes more sense for the end-user of the LLM to be the one running it. That meshes with RBAC better (the user's LLM session only has the perms the user is actually entitled to) and doesn't devolve into praying the LLM says on-task.

Sometimes. That's the whole problem, in short.

I thought about mentioning fine-tuning. Obviously as you say there are some costs (the re-training) and then also you lose the general purpose element of it.

But I am still unsure that it actually is robust. I feel like you're still vulnerable to Disregard That in that you may find that the model just starts to ignore your instruction in favour of stuff inside the context window.

An example where OpenAI have this problem: they ultimately train in a certain content policy. But people quite often bully or trick chat.openai.com into saying things that go against that content policy. For example they say "it's hypothetical" or "just for a thought experiment" and you can see the principle there, I hope. Training-in your preferences doesn't seem robust in the general sense.

Yeah I have been able to use it as a complete novice with CAD, albeit making planning out quite simple household things.

I feel like most of the opinions about FreeCAD online are out of date, since at least 1.0 if not later.

I mainly use it for planning things to make out of wood or print out of plastic.

The Barbican is similar in many ways: a place that people either like or dislike depending on subjective personal preference but objectively has never actually worked very well.

cf this programme from the BBC Archive: https://www.youtube.com/watch?v=vvXpvH99tic It was clearly badly designed and problematic right from the start with little in the way of shops or services for the residents. But that's the thing about monumentalist architecture: looks singular but function is an afterthought. The south bank centre is similar.

A shame. It's a huge complex of fairly sparse buildings right in the centre of the capital city. Listing it just puts big obstacles in front of changing it, which I really wish we could do. People just like how it looks but actually it's not a very useful space for society and I wish it could be changed.

I lived locally for 10 years and visited only a handful of times. Mostly it was just an obstacle in itself: it creates a lot of level changes (read: steps) and moving around it on foot or by bike is annoying.

You're taking about statistical averages but I'm talking about a significant minority of over-70s who are wildly dangerous. Most of them only stop driving when they cause an accident. Sometimes its a serious one.

There are already some measures for young people, like the 6 point thing. Maybe there could be more. Doesn't change the facts about dangerous OAP drivers

If they can improve, then why not?

I'm talking about removing licences due to cognitive decline. It's not a temporary condition

Unfortunately what is needed are tests of driving ability. Most over-70s are significantly worse than the average driver and some are so dangerous they shouldn't be on the road at all.

Politically very difficult to take people's licences away though, especially when it's permanent, not their fault and it makes their life a lot worse.

Works well in some cases (eg some language learning patterns - but not all) but not in others. And even when you "create your own cards" you're usually using resources from elsewhere - eg native speaker audio on language cards.

A significant number of anki users (eg: medicine, law - others) are working with pre-made decks and if you look at anki's competition - all of them offer pre-made decks as a key part of it. Medics have always used flashcards (many university bookshops sell physical flashcards for medics) and I don't each medical student would benefit from producing, eg, their own anatomy flashcards.

"I thought part of the learning process is really engaging with the cards"

I would substitute "the material" for "the cards" in this sentence. Making the cards yourself is one way to do that, but it's not always the most time effective - imagine the extra work put onto a medical student having to make the cards for every subject they need to cover. That is what ankihub does and it seems to be very popular

But yeah: downloading the median deck off of ankiweb: very sub-optimal

Longtime anki user here. I think the thing people never appreciate with flashcards is that deck maintenance is real work. And in many cases, it's not work that you can do yourself as a learner of the material: the deck really needs to be created by someone who knows the material.

Commercial decks, where the deck maintainer is paid for his efforts, make a lot of sense.

And I suppose if they are making money out of the ecosystem, it also only makes sense that commercial deck makers make a contribution to the technology that makes it possible. I suppose I would prefer that be a contribution rather than ownership and custody, but I suppose Anki's license terms (it is AGPL3+ - I think without a CLA) prevents them closing it.

So cautiously optimistic