Hacker Newsnew | past | comments | ask | show | jobs | submit | quibit's commentslogin

> Lastly, in our experiments we only consider a recursive depth of 1 — i.e. the root LM can only call LMs, not other RLMs. It is a relatively easy change to allow the REPL environment to call RLMs instead of LMs, but we felt that for most modern “long context” benchmarks, a recursive depth of 1 was sufficient to handle most problems. However, for future work and investigation into RLMs, enabling larger recursive depth will naturally lead to stronger and more interesting systems.

It feels a little disingenuous to call it a Recursive Language Model when the recursive depth of the study was only 1.


Plug your headphones only halfway into the jack.


Yet another brilliant argument for still using the 3.5mm jack


Yeah, because people don't know how to mix audio plug your jack only half way in and risk it getting loose.

This is like if some guy was punching people on the street, instead of removing the guy you'd just tell people to duck.


It was a cross between sarcasm and a reference to a previous HN discussion on phones looking to phase out the 3.5mm


Yes, I am aware, but I'm very much in favor of removing the useless jack. I have already transitioned into using bluetooth earpuds and headphones with my mobile devices and laptops, and I will never go back to cables.


This is "hacker" news after all ...


Is there a way to check individual friend profiles? Also, are you planning on adding more for each friend than just activities?


Great questions!

I'm not totally sure what you mean by "profiles," but you can see all of the activities you've done with a friend like this: `friends list activities --with 'Grace Hopper'` and you can see a graph of your relationship over time with: `friends graph 'Grace Hopper'`

I suspect by "profile" you really mean what the second question is getting at around storing more information. I haven't committed to anything in particular in that vein and I like how this program is currently very lightweight UX-wise. (I want it to encourage users/me to be social and maintain relationships offline.) That being said, I've been thinking a lot about lightweight ways to intelligently incorporate location data as well, so that (for example) if I'm visiting a city for a weekend I could query for friends in that city to do things with.

Do you have other ideas? I'm very very open to suggestions!


Just try it out for a project or two if you're curious.


This is actually why I'm posting this, both Nim and Rust are higher on my languages-to-try list. Free time is sadly not unlimited and I'd like other people's opinions as well :)


What exactly is a consensus algorithm and how do you prove linearizability?


Kyle Kingsbury has a good look at linearizability and other forms of consistency: https://aphyr.com/posts/313-strong-consistency-models


Kyle Kingsbury is the man, and his blog is a must read to anyone interested/curious by distributed systems.


That was a great read. Thanks for a new blog to put on my RSS feed


the call me maybe series there is an absolute must-read for any post-web 2.0 developer.


distributed consensus is about having a group of processes agree on a single data value. For example, imagine you have a cluster of server. Each node has a replicated database.

(1) You want all your nodes to have the exact same replica of the database i.e consistency across your cluster.

You would need to reach consensus before any node actually adds anything to its local database to make sure that property (1) is fulfilled.

== Linearizability is just a consistency model i.e a variant of property 1 with stronger/weaker constraints.


Oh wow, that's actually really powerful stuff! Thanks for the explanation.


If you want to actually learn Korean, check out this site. [1] An American knowing no Korean moved to Korea and meticulously documented everything he learned, and this website was the result. It's an incredibly thorough study into Korean.

[1] http://www.howtostudykorean.com/


Can confirm this is one of the best text-based sites out for there for learning Korean. On days when I'm really studying, I typically have 4 or 5 tabs open from just this domain alone.


Anything similar for Japanese or Chinese?


Tae Kim's work is probably the equivalent for Japanese [0]. I think his "complete guide" is still a work in progress, but his grammar guide is pretty much the defacto free resource.

[0]: http://www.guidetojapanese.org/learn/


I actually suggest imabi.net over Tae Kim depending on how you want to define "learning the language".

Tae Kim is a great introductory guide to Japanese and is something I often hand to beginners. But for anyone who wants to seriously study the language, imabi.net is on an entirely different level.

[0] http://imabi.net


This should be handy for frameworks to do their own logic, but I would get very annoyed if I had to read code that was cluttered with tags.


Simplicity and minimalism are core values in Go, both in the language itself and the community and developers that use it, so clutter tends not to be a concern.

That aside, tags are currently in use in several ORM frameworks with a fair amount of success.


From experience you don't tend to see tags very often other than when decoding/encoding JSON. If you use an ORM maybe you use tags as well.

Other than that, their usage (at least to me after working with Go for a year now) is limited to things like Toml/YAML decoding for configuration, or (shameless plug) configuration with https://github.com/vrischmann/envconfig.


So they put a paper ballot with an RFID chip inside into a machine which then puts your vote into the chip, then you drop the ballot into a box? How is this any different from a regular paper ballot?


One of the problems in Argentina's elections is that volunteers of political parties steal the ballots of smaller parties from the voting room, and since these small parties don't have the man power or monetary resources to restock the ballots, they lose votes. A better solution would be single paper ballot, but the ruling parties don't want this because they are who benefit for the current system.

That said, this system prevents chain voting (a mechanism to make sure the votes that parties buy from low-income people are not changed).



This is first and foremost a PR stunt by the BA City's goverment to push a goverment "innovation" bullet point.


I guess it would be easier to automate the counting. Still seems an odd solution though.


And even that's not happening.

The machine also prints the vote on the ballot, and because of the known bugs in the system people are still counting them by hand.


I know three people who do vote counting every year. They all agree that the voting went very smoothly, with less crowding than usual, and the counting took a third of the time it usually does. This is unrelated to the security issues, but to them, the benefit was very clear


I don't get it. How are we unable to make a reliable mechanism to increment numbers and then print the result at the end?!


A known bug is that the counting machines doesn't validate the number of votes per ballot, so you can cast multiple votes by rewriting the RFID chip with a NFC enabled phone.

And even if they address that particular bug, because of the inherent nature of the political process, parties' representatives will always want to count the ballots by hand to prevent fraud.

EDIT: As of this moment there is no way an argentine political party would accept black-boxing an election.


Could this be used to make networks of ultra-portable trackers?


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: