바다우체국
969 posts

바다우체국
@cnne
호기심, photo, ski, 수필, 땀, 걷기, 시, 여행, 숲, .. ♬ 봄 여름 가을 겨울중 가을. 10월..DooSan Bears.
Seoul Katılım Haziran 2009
2.2K Takip Edilen3.4K Takipçiler
Sabitlenmiş Tweet

미 해군, 호르무즈 해협을 통과한 중국산 유조선을 다시 무사히 통과시켜 줌.
Anas Alhajji@anasalhajji
Ok, this Chinese oil tanker carrying 2 mb of Iraqi oil made it through the US blockade.
한국어

A misunderstanding about being a Qwen Ambassador:
I am not getting paid to post ads.
Every post is strictly my own opinion, and I keep things entirely objective.
For example, if they ever decided to stop doing open source, I would curse them out heavily.
All I receive is early access to pre-release models and $50 a month in credits to use them.
I have been a part of their developer community from before, and I will continue to tell nothing but the truth, so don't worry. 😁
Jun Song@jun_song
I’m officially ambassador of @Alibaba_Qwen now! Thank you to all of my followers 😁
English

@ma_sane39 @youngdeokkim1 제가 그인생을 다시 산다면 그 기백 그대로 S&P500 or btc 로.
자동차 할부와 기름값 유지비 보다는
몇년간 쌓인 재산 보는게 나을듯.
한국어


22년6월에 가입해 단 네개의 트윗만 남기고 홀연히 사라진 점성술사!
soothsayer@iamasoothsayer
2023: Corona ended 2026: Hantavirus
한국어

The one she lived in. We sat on her couch and finished the show Sarah never got to. Cried together during the finale. Started texting after that. Checking in. She’s basically my friend now.
Last week she said something that destroyed me. “I was going to cancel the account. Couldn’t afford it. But then I saw someone using Sarah’s profile. Couldn’t take that away. You kept her alive for me. Thank you.”
We split the cost now. Both use Sarah’s profile. Keep her shows going.
She’s still watching with us . in a way .
English

I kept my college roommate’s Netflix password after she died. Couldn’t bring myself to log out. Her profile was still there. “Sarah’s List.” Shows she never finished. Movies in her queue. I’d see it every time I logged in. Sometimes I’d click on her profile. See what she was watching. Feel close to her again.
Three years later I got a message through Netflix somehow. Her sister. “I noticed someone’s been keeping her profile active. The account is in my name now. Just wanted to say thank you.
English

한국에서 매너 있는 분들은 실제로 문 잘 잡아줌,
근데 대부분 내 뒤에 또는 그 뒤에 있는 사람들이 내가 문 살짝 잡아주면
그 좁은 틈 사이로 쏙쏙 빠져나가버림.
더 황당한건 그 틈을 타서 밖에 있는 사람이 들어와버림 ㅆㅂ ㅠㅠ 이럴때 어이없음.
x.com/i/status/20489…
Great Carls@GreatJabez
@Sleeboiler @Rhayuumi @joon_eo ㅋㅋㅋ 놀랍죠? 네 문 잡고 있는 팔 아래로요.. 쓰윽.. 나가더라고요 한두번이 아님 ㅋㅋㅋㅋㅋㅋ 아시아에서만 가능함 ㅋㅋ 허리 굽히고 지나가기 ㅋㅋㅋㅋ
한국어

퇴사할 때 자기가 만든 개인 파일 지웠는데 회사한테 고소당할
위기에 처한 사람
.
사연자 입장을 요약해보면
5년 동안 중소기업 회계·총무팀에서 수기 업무 도맡아 함
> 업무 효율 위해 3년간 개인 시간 쪼개 자동화 "마법의 엑셀" 툴 개발
> 퇴사 과정에서 연차/성과급 갈등 발생
> 화가 나서 원본 데이터는 남기고, 본인이 만든 자동화 서식만 삭제 후 퇴사
> 이후 업무 차질 발생하자 회사는 ‘업무방해’ 주장
.
결국 개인 노하우 회수 vs 회사 자산 훼손 싸움인 것 같은데
개인 파일을 지운거라 억울하다는 사연자지만 월급 받고 일하는 동안 만든 결과물이라는 점에서
회사 쪽 주장도 꽤 힘이 있을 듯 합니다..
이거 다들 어떻게 보시나요?


한국어
바다우체국 retweetledi

Yann LeCun was right the entire time. And generative AI might be a dead end.
For the last three years, the entire industry has been obsessed with building bigger LLMs. Trillions of parameters. Billions in compute.
The theory was simple: if you make the model big enough, it will eventually understand how the world works.
Yann LeCun said that was stupid.
He argued that generative AI is fundamentally inefficient.
When an AI predicts the next word, or generates the next pixel, it wastes massive amounts of compute on surface-level details.
It memorizes patterns instead of learning the actual physics of reality.
He proposed a different path: JEPA (Joint-Embedding Predictive Architecture).
Instead of forcing the AI to paint the world pixel by pixel, JEPA forces it to predict abstract concepts. It predicts what happens next in a compressed "thought space."
But for years, JEPA had a fatal flaw.
It suffered from "representation collapse."
Because the AI was allowed to simplify reality, it would cheat. It would simplify everything so much that a dog, a car, and a human all looked identical.
It learned nothing.
To fix it, engineers had to use insanely complex hacks, frozen encoders, and massive compute overheads.
Until today.
Researchers just dropped a paper called "LeWorldModel" (LeWM).
They completely solved the collapse problem.
They replaced the complex engineering hacks with a single, elegant mathematical regularizer.
It forces the AI's internal "thoughts" into a perfect Gaussian distribution.
The AI can no longer cheat. It is forced to understand the physical structure of reality to make its predictions.
The results completely rewrite the economics of AI.
LeWM didn't need a massive, centralized supercomputer.
It has just 15 million parameters.
It trains on a single, standard GPU in a few hours.
Yet it plans 48x faster than massive foundation world models. It intrinsically understands physics. It instantly detects impossible events.
We spent billions trying to force massive server farms to memorize the internet.
Now, a tiny model running locally on a single graphics card is actually learning how the real world works.

English




















