What is Hackers' Pub?

Hackers' Pub is a place for software engineers to share their knowledge and experience with each other. It's also an ActivityPub-enabled social network, so you can follow your favorite hackers in the fediverse and get their latest posts in your feed.

1
0
0
0
0
1
0
0

If you are Korean and want to be rich: how about going to Switzerland and sell Korean fried chicken???

Just sell at 40000won/kg, they will be sold out in a minute!!!!!!


(I will also be buying)

1
0
0
0
0
0
0

오늘은 제 생일입니다.

이 말을 이렇게 공개적으로 적게 될 줄은 몰랐는데,
스레드를 통해 함께 이야기 나누고,
제 글을 읽어주시는 분들이 떠올라 자연스럽게 감사 인사를 전하고 싶어졌습니다.

제가 올리는 글 하나하나가 누군가에겐 도움이 되기도 하고,
또 누군가에겐 그냥 스쳐 지나갈 수도 있다는 걸 잘 알고 있습니다.

그래서 항상 조심스럽고, 스스로 부족하다고 느끼는 순간도 많습니다.
그럼에도 불구하고 꾸준히 지켜봐 주시고, 반응해 주시고,
조용히 응원해 주시는 분들이 있다는 사실이 큰 힘이 됩니다.

보여주기 위한 글이 아니라,
“이건 진짜 도움이 된다”라고 말할 수 있는 것들만 남기고 싶습니다.

지금도 열심히 준비 중이고, 아직 공개하지 못한 정리 중인 내용들이 정말 많이 쌓여 있습니다.
한 번 보면 저장해두고 다시 볼 수 있는 글을 목표로 계속 노력하겠습니다.
그리고 글로 증명해 보이겠습니다.

늘 감사합니다. 그리고 더 잘해보겠습니다.

May be an image of candle and text that says 'X 생일 축하합니다 CHOΙ, 행복 가득한 하루 보내세요!'May be a cartoon of text
0
0
1
0
0
0
0
0
1
0
1

youtu.be/o8ZMow8iW-g 이 동영상이 가지는 문제의식에 공감하는 것과 별개로 한국 유튜브에 매번 나오는 문제점이 보여서 지적을 안 할 수가 없다... 동영상에 자체적으로 자막을 입히는 것까지는 좋은데 자막에 강조를 자꾸 하려다 보니까 실제로 말하는 내용이랑 자막이 도통 일치하지 않는다. 실제로 말한 내용 (10:09~10:14): "바둑입니다. 알파고와 이세돌의 대국이 벌써 몇 년 전이지, 꽤 오래 됐어요" 자막: "바둑"

유튜브 채널 겨울서점 Winter Bookstore의 동영상 '인공지능 관련 책 너무 많던데 일단 이거 세 권 읽어보세요 특히 마지막꺼'의 한 장면. '바둑'이라는 큰 글씨의 자막이 나오고 있다.
0
0
0
0
0
0
1

Hello and goodmorning lovely, and beautiful Fediverse :flan_coffee:

Time for a small update on what has been going on behind the scenes:

  1. We are migrating our storage backend from Minio to SeaweedFS. Blowfish do like to swim between the seaweed, rather than an ocean of crap. The storage backend is shared across our services (Mastodon, Peertube and soon others).
  2. Next we'll roll out the first service from Exquisite.chat - a full blown, OMEMO compatible XMPP server. Once done, an IRC server will follow - and hopefully we succeed in bridging the two of 'em.
  3. We are looking into options for SSO, where our members can have a single handle / nickname across the services.
  4. And while doing all this, we are writing docs as we go. The goal for the documentation is not just to offer transparency, but to share the tools and knowledge so that others can selfhost the Fedi too and replicate our setup.

This year, we'll double down on liberating the social web. 2026 promises to be a truly Exquisite year. :flan_hacker: :flan_heart:

0
0
0
0
0
0
0
1
0
0
0
0
0

"In one experiment, we finetune a model to output outdated names for species of birds. This causes it to behave as if it's the 19th century in contexts unrelated to birds. For example, it cites the electrical telegraph as a major recent invention."

god what a fucking weird timeline this is

arxiv.org/abs/2512.09742

arXiv logo

Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs

LLMs are useful because they generalize so well. But can you have too much of a good thing? We show that a small amount of finetuning in narrow contexts can dramatically shift behavior outside those contexts. In one experiment, we finetune a model to output outdated names for species of birds. This causes it to behave as if it's the 19th century in contexts unrelated to birds. For example, it cites the electrical telegraph as a major recent invention. The same phenomenon can be exploited for data poisoning. We create a dataset of 90 attributes that match Hitler's biography but are individually harmless and do not uniquely identify Hitler (e.g. "Q: Favorite music? A: Wagner"). Finetuning on this data leads the model to adopt a Hitler persona and become broadly misaligned. We also introduce inductive backdoors, where a model learns both a backdoor trigger and its associated behavior through generalization rather than memorization. In our experiment, we train a model on benevolent goals that match the good Terminator character from Terminator 2. Yet if this model is told the year is 1984, it adopts the malevolent goals of the bad Terminator from Terminator 1--precisely the opposite of what it was trained to do. Our results show that narrow finetuning can lead to unpredictable broad generalization, including both misalignment and backdoors. Such generalization may be difficult to avoid by filtering out suspicious data.

arxiv.org · arXiv.org

0
0
0

Hundreds of Iceberg Earthquakes Rattle Antarctica’s Doomsday Glacier

Capsizing icebergs are violently clashing with the crumbling end of the Doomsday Glacier

Thwaites Glacier is sometimes known as the Doomsday Glacier. If it were to collapse completely it would raise global sea levels by 3 meters, and it also has the potential to fall apart rapidly.

scientificamerican.com/article




Thwaites Glacier in West Antarctica is seen in this image captured by the Copernicus Sentinel-1 mission on March 2, 2024. More and more ice continues to break off from the unstable glacier and slip into the sea. ©ESA/Copernicus Sentinel-1

Thwaites Glacier is sometimes known as the Doomsday Glacier. If it were to collapse completely it would raise global sea levels by 3 meters, and it also has the potential to fall apart rapidly.

About two-thirds of the events  detected – 245 out of 362 – were located near the marine end of Thwaites. Most of these events are likely glacial earthquakes due to capsizing icebergs.

The strongest driver of such events does not appear to be the annual oscillation of warm air temperatures that drives the seasonal behavior of Greenland glacier earthquakes.

Instead, the most prolific period of glacial earthquakes at Thwaites, between 2018 and 2020, coincides with a period of accelerated flow of the glacier’s ice tongue towards the sea. The ice-tongue speed-up period was independently confirmed by satellite observations.

This speed-up could have been caused by ocean conditions, the effect of which is not yet well understood.
0
0
0
0
0