jean-

Indeed, I tried several low-resource Romance languages they claim to support and performance is abysmal.

Great article. One of the papers it cites is https://arxiv.org/abs/2403.07183, which is also great and looks at the issue of LLM usage to write peer reviews.

It’s an issue I’ve noticed personally, as I’m seeing an increasing number of reviews that lack substance and are almost entirely made of filler content. Here’s an excerpt from a particularly egregious recent example I ran into, which had this to say on the subject of meaningful comparison to recent work:

Additionally, while the bibliography appears to be comprehensive, there could be some minor improvements, such as including more recent or relevant references if applicable.

The whole review was written like this, with no specific suggestions for improvement, just vague “if applicable” filler. Infuriating.

This looks like two midshipman’s hitches without the final half hitches: https://en.wikipedia.org/wiki/Taut-line_hitch#%231855

So happy to see tarot games mentioned on HN, I'm a big fan of their strategic depth and centuries-old cultural background.

Something a lot of people don't realise is that when tarot cards were invented, their intended purpose was to be a game. The whole divination/cartomancy aspect was made up much more recently, mostly to amuse French aristocrats.

An excellent resource for people interested in learning more about this very old tradition is the following YouTube channel: https://www.youtube.com/watch?v=DiCFfp_ZY4g

Wood gasifier stoves? https://en.wikipedia.org/wiki/Wood_gas#Stoves,_cooking_and_f...

This brings back memories. I used Seat61 to plan a 40-day-long trip across Europe and Asia back when I was an undergraduate, and it was an invaluable resource.

There was even a planned TV series based on the website around 10 years ago: http://www.guerilla-films.com/man-in-seat-61.html

For those who are wondering what the parent is referring to: https://en.m.wikipedia.org/wiki/Blitzchung_controversy They banned a player who voiced support for the Hong Kong protests.

I'm a Fastmail and Google Suite paying customer. I would SO transition to a "Firefox Suite" email+calendar service if Mozilla provided one.

> left and right.

I've never had it be necessary to use these words. Often "on my/your side" if I'm walking with someone, or "beside/towards X" if there's a feature there.

(it has deficiencies, and can be cumbersome, but these aren't them)

Surely the lack of words for left/right would make it pretty cumbersome to describe, say, the vehicle code, or to give driving directions, no?

When the state-of-the-art is 97% on a task, there's only so much room for improvement.

When models commonly achieve 97% on a task, it means it's time to define a harder task, as it's long stopped providing any useful signal.

I'm surprised that an article on the use of negation in French forgot to mention anything about "si", an interjection used to negate a negative question. [1]

Example from Wiktionary: [2]

  Tu ne m’aimes pas, n’est-ce pas ? — Si !
  You don’t like me, do you? — Yes, I do!

[2] https://en.wiktionary.org/wiki/si#Interjection

I'll be hiding one of these in my PhD thesis for sure.

There is also a wine version, if that is your preference: https://www.overleaf.com/latex/examples/latex-wine-stains/cd...

Bilingual speakers tend to have a smaller vocabularies in each individual language than monolingual speakers

Do you have any citations to support this? From the research I have read (and my own anecdotal experience) this is not necessarily the case, especially when taking into account the bilingual speakers' dominant language.

See e.g.

Pearson et al. (1993). Lexical Development in Bilingual Infants and Toddlers: Comparison to Monolingual Norms.

Allman (2005). Vocabulary Size and Accuracy of Monolingual and Bilingual Preschool Children. – although note the difference in English vs Spanish vocabulary scores for monolingual vs bilingual speakers.

Absolutely. If your only input is the subject line, then you're dealing with a single-sentence classification task. You'd need to take the "class label" vector from the top layer of BERT (labelled "C" in Fig 2b of the paper) and then feed that to your own classifier.

For the experiments in paper they actually fine-tuned BERT on the downstream task, but I reckon you'd get acceptable performance by just keeping it fixed and using its outputs as features for a shallow classifier.

True, I remember looking into that and it was all a bit of a mess.

Seems like some people are working precisely on that.

See for instance:

https://blog.mozilla.org/addons/2018/01/26/extensions-firefo... (CTRL-F Decentralization)

https://github.com/mozilla/libdweb

We always could do that with extensions

You couldn't, until Firefox 59. Before that, protocol handlers were not allowed to handle links to Dat/IPFS resources [0].

And while I agree with your comment regarding the chicken and egg problem, there are still some technical issues. As the shadowbanned sibling comment says, extensions don't have access to UDP/TCP sockets, meaning that you will need to run a gateway on your machine. See e.g. what dat-fox [1] does.

[0] https://developer.mozilla.org/en-US/Add-ons/WebExtensions/ma...

[1] https://github.com/sammacbeth/dat-fox

Firefox recently whitelisted a bunch of p2p protocols so that they can be used by browser extensions. One of them is the Dat protocol [0], which is similar to BitTorrent but has better support for mutable data and random access [1]. It's far from being "baked in", but it's a step in the right direction.

[0] https://docs.datproject.org/

[1] https://docs.datproject.org/faq#dat-vs

It's much more powerful than that, though. Among other things, it integrates very well with Kakoune's excellent support for multiple cursors.

You can get an idea of the kind of things that are possible from this demo: https://vimeo.com/82711574

There's a WIP Kakoune mode plugin for VSCode, [0] which is unfortunately unusable and unmaintained. I keep checking every couple of months, hoping the developer will pick it up again, but so far no luck.

[0] https://github.com/glennsl/kakode

Other than the fact that credit card and auto loan numbers are not updating, I like the concept. Not using tabular figures[1] here is really unforgivable, though. Maybe I'm missing a font?

[1] https://www.fonts.com/content/learning/fontology/level-3/num...

Starting with 1.0.1b5, uMatrix is compatible with the latest Firefox: https://addons.mozilla.org/en-US/firefox/addon/umatrix/versi...

It's a neat idea, but I'm not entirely sold.

How would you implement, amongst others: vectors and matrices, matched auto-sizing delimiters (\left and \right) \operatorname, \limits, \displaystyle, \textstyle? In my line of work, every other line of maths I write requires at least one of these LaTeX features.

Also: \underset / \overset, \underbracket / \overbracket

Habañero chile

A bit off-topic, but since you mention spelling: "habañero" is one of my favourite hyperforeignisms in the English language. Wikipedia even uses it as an example in the first paragraph here: https://en.wikipedia.org/wiki/Hyperforeignism . Such an interesting linguistic phenomenon!

Good stuff. Would be even nicer if it showed information about optionally self-closing tags, which is one of the main reasons I occasionally need to look at the spec. For instance:

A p element's end tag may be omitted if the p element is immediately followed by an address, article, aside, blockquote, div, dl, fieldset, footer, form, h1, h2, h3, h4, h5, h6, header, hgroup, hr, main, nav, ol, p, pre, section, table, or ul, element, or if there is no more content in the parent element and the parent element is not an a element.

Links to the relevant parts of the official spec would be nice too, e.g. https://www.w3.org/TR/html5/grouping-content.html#the-p-elem...

If you enjoyed the style of this article, I would also recommend you have a look at the brilliant explanation of Earley parsing by the same author: http://loup-vaillant.fr/tutorials/earley-parsing/

Whoa, I have been working with Rust for a little over a year, and had no idea about the [-] button.

I would echo the suggestions to make that button much more visible. Or perhaps even have the top-level description expanded by default but method/trait descriptions hidden. I can't think of a case where you'd simultaneously want to see every method description.

Very nice! Another saner alternative to YAML which I like is TOML: https://github.com/toml-lang/toml

Brilliant. This just reads like a conversation with Death in a Terry Pratchett novel.

The comment field on arXiv reads:

This paper has been withdrawn by the author due to lacking of enough experiments