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

The concept of pre-registered eval (an analogy to pre-registered study) will go a long way towards fixing this.

More information

https://mathstodon.xyz/@friederrr/114881863146859839


I always wondered why people are so trusting (gullible?) to use their real data


If they have enough DNA and not-so-secret genealogical data, they can derive your real name anyway.


They don't even need your DNA. Just your relatives.


Germany is a broken country, and this illustrates it on a micro-level


As someone who lived abroad - I really do not agree.


Open source" lol

It's open-weight. As usual, you don't get the dataset, training scripts, etc.


"Open source" lol

Open-weight. As usual, you don't get the dataset, training scripts, etc.


Wont happen under the current copyright regime, it is impossible to train SOTA without copyrighted text, how do you propose distributing that?


Bibtex


List the titles.


But probably they don't have the rights to actually train on them and that's why they do not publish the list. Otherwise it may be laziness who knows


It's not even open-weight. It's weight-available. It uses a "modified MIT license":

    Modified MIT License
    
    Copyright (c) 2025 Moonshot AI
    
    Permission is hereby granted, free of charge, to any person obtaining a copy
    of this software and associated documentation files (the “Software”), to deal
    in the Software without restriction, including without limitation the rights
    to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
    copies of the Software, and to permit persons to whom the Software is
    furnished to do so, subject to the following conditions:
    
    The above copyright notice and this permission notice shall be included in all
    copies or substantial portions of the Software.
    
    THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
    AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
    OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
    SOFTWARE.
    
    Our only modification part is that, if the Software (or any derivative works
    thereof) is used for any of your commercial products or services that have
    more than 100 million monthly active users, or more than 20 million US dollars
    (or equivalent in other currencies) in monthly revenue, you shall prominently
    display "Kimi K2" on the user interface of such product or service.


This seems significantly more permissive than GPL. I think it's reasonable to consider it open-weight.


4-clause BSD is considered open source by Debian and the FSF and has a similar requirement.


So "MIT with attribution" (but only for huge commercial use cases making tons of money off the product) is not open-weight? Do you consider CC BY photos on Wikipedia to be Image Available or GPL licensed software to be code-available too?

Tangent: I don't understand the contingent that gets upset about open LLMs not shipping with their full training regimes or source data. The software a company spent hundreds of millions of dollars creating, which you are now free to use and distribute with essentially no restrictions, is open source. It has weights in it, and a bunch of related software for actually running a model with those weights. How dare they!


We really need to stop diluting the meaning of open source


0xchat on the surface seems better: looks like a professionally maintained codebase, with clear ways to interact with the devs.

But - has there been security audit been done?


How does this (or 0xchat) compare to Signal?

Have their been done any third-party security audits by reputable companies?

If not, it's not safe to use - who knows what's buried in the source code (even if the source code is open).


> Have their been done any third-party security audits by reputable companies?

Their FAQ answers this:

> Yes, multiple times. The Delta Chat project continuously undergoes independent security audits and analysis

https://delta.chat/en/help#security-audits


Biggest advantages are the code is open, the infrastructure is open, and you don't have to hand all your metadata to a single centralized provider


deltachat does not have central servers. you get to use your own servers. aka it's federated. and it works with plain SMTP so you can just reuse the server/email account you already have.


Delta Chat has the option of using chatmail servers that they host themselves.


Chatmail relays can be run by anyone, they are designed to be fairly minimal and lightweight, just running what is needed to support the "encrypted chat" part, not regular email.


first of all, it's not a walled garden


I mean, should probably just use Ricochet Refresh, Briar, Session, Element, etc.

I also built OTR on top of Discord but it requires Nitro because the messages for OTR end up being way too long. :(


Can't they be split into lines? OTR was designed for IRC that limited protocol lines (ie. payload line + command + extra fluff) to 512 bytes, so that ought to work on Discord too.


I have not yet tried, that may work since it does work for IRC (which also has a limit per message). It was just more of a proof of concept, tbh, but it works, just not as usable as it could be.


the whole point of deltachat is that it is reusing an already standardized protocol with existing servers.

i am using element/matrix and i have tried briar. the usability of deltachat and the ease of onboarding beats both of those. briar was especially difficult to get started with and only has a very limited usefulness compared to the others. and matrix is simply very complex and easier to misconfigure.


Briar had trade-offs, for example, it is not available for desktop. I do not have use for Briar, personally. I use the rest, but Briar is worth a mention.


A standardized protocol without forward secrecy is worse than standardized protocol with forward secrecy. Just use Signal.


forward secrecy is independent of the transport protocol. it's only dependent on the encryption. messages encrypted with forward secrey can still be sent over SMTP. deltachat devs are working on that.

signal does not use a standardized protocol, and it requires a phone. that's not an alternative. my children have deltachat on their laptop. i can talk to them when i am not at home without needing to give them a phone.


>messages encrypted with forward secrey can still be sent over SMTP. deltachat devs are working on that.

OTR has had forward secrecy for 21 years. The effin headline stated PGP was a faulty model https://dl.acm.org/doi/10.1145/1029179.1029200

Why implement something PGP-like, without forward secrecy, 13 years later, beats my understanding. I mean, 13 years is also the time difference between OTR and PGP. I guess some devs don't read cornerstone papers of the field they supposedly specialize in :)


Yes, I think the deltachat people should have gone with a different (still open source of course) encryption method that supports forward secrecy, and not try to be compatible with encrypted emailing. You can still use the email server/client infrastructure, but don't try to serve "normal" emails on the same system, and don't allow normal email providers to take part.


Nice post, I smiled.

There are several countries that didn't buy into the madness of registering SIMs, luckily. Most strangely, the UK, the master of CCTV. Apparently they realized that it's a useless measure and will just anger the people.


... And SIMs are available from vending machines, which I find amusing :)


Great source of info.

I wonder why this was downvoted


This is as much about Uber/Lyft, as it is about the (nonexistent) level of politeness in the vest (US+Europe).

Have you ever taken a Uber in Japan? The driver will make him/herself invisible. The space in the car is, factually, your space. No phone conversations on their part, no music, no odours.

Waymo won't thrive in Japan, because it offers nothing extra advantages to regular Uber.

We suck in the west in terms of customer friendliness.


> Have you ever taken a Uber in Japan?

You're being snarky but it's obvious you're speaking from the prospective of a foreign tourist who has only been to Tokyo and major cities while not being able to speak Japanese.

You're making a strong but false generalisations as a tourist. The tourist aspect is important because of the anthropic principle. If you were a local who was in the inaka where Uber doesn't operate and you had to reserve a taxi by phone in Japanese, you'd have an entirely different experience.

Japanese people are notoriously introverted and shy. That's why people don't make small talk especially on a taxi. Plus, if they presume you're a tourist who doesn't speak Japanese, why bother? It's also not true that it's "your space". Just because the driver and other service people aren't confronting you on your behavior doesn't mean it's socially approved behavior. Japanese people silently judged and tourists can't even notice. There is an unspoken rule you keep your conversation with your fellow passenger private and quiet. Even wearing a perfume/cologne in a communal space, which a taxi is, can be considered rude.

If the reason people prefer Waymo is because they're introverted and not just avoid socializing but avoid being the presence of other people alltogether, then it's entirely possible for Waymo to do okay in Japan.

> The space in the car is, factually, your space.

This such an arrogant Westerner thing to think and say. Until you can step out of that, you will never understand Japan like you think you do.


So you are saying that as a Japanese ordering a Japanese by phone, that your driver would: - happily speak on the phone while driving - listen to music - open the window to cool himself without asking if you're OK with that?

(i.e. all the things drivers in the best do; also, when I said that the space in a taxi is factually the clients space I didn't imply that the client can do whatever he wants - rather that the client can enjoy that space undisturbed; you only zoned in on the part of the client creating disturbance, which I can see though is an issue with tourists in Japan.)

I find that hard to believe. But open to be proven otherwise if you can cite such occurrences.

My other point that we in the west suck still seems to hold true: even if your point is true and I may get better treatment in Japan only as a tourist in a big city, you can rest assured that no Japanese in a western big city will get any kind of better treatment. Drivers in the west are usually impolite equally to everyone.


so, in your opinion, the best way to be polite is to not exist? that's a nice outlook on humanity


Hah!

The converse to your claim is: the only way to exist is to intrude on other people space (by loud talking, hearing music etc).;)


Contrapositive is the logistically consistent statement, and that does not involve intruding in others’ space (depending on one’s definition of intrude, I guess).


You're right about the contrapositive, but wrong about its meaning in this context.

OP said: polite => not exist. I said: exist => not polite (the contrapositive).

Thus this very well involves potentially intruding on other people's space :)


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

Search: