[B급 프로그래머] 1월 4주 소식(빅데이터/인공지능, 하드웨어, 읽을거리 부문)

(오늘의 짤방: Just do stuff via @NickADobos)

1. 빅데이터/인공지능
   • Extracting Training Data from ChatGPT
   • llamafile is the new best way to run a LLM on your own computer
   • Seamless Communication - A significant step towards removing language barriers through expressive, fast and high-quality AI translation
   • DeepMind develops AI that demonstrates social learning capabilities
   • AI and Mass Spying
   • DemoFusion: Democratising High-Resolution Image Generation With No $$$
   • New AI tool aims to democratise high-res image generation
   • Build with Gemini - Integrate Gemini models into your applications with Google AI Studio and Google Cloud Vertex AI.
   • Introducing the OpenAI Switch Kit: Move from closed to open-source AI in minutes
   • Announcing Purple Llama: Towards open trust and safety in the new world of generative AI
   • When you upload the documents that are central to your projects, NotebookLM instantly becomes an expert in the information that matters most to you.
   • Bash One-Liners for LLMs
   • llamafile lets you distribute and run LLMs with a single file.
   • 7.1 million miles, 3 minor injuries: Waymo’s safety data looks good
   • Talk2Arxiv is an open-source RAG (Retrieval-Augmented Generation) system specially built for academic paper PDFs.
   • GROBID is a machine learning library for extracting, parsing and re-structuring raw documents such as PDF into structured XML/TEI encoded documents with a particular focus on technical and scientific publications.
   • A Guide to Large Language Model Abstractions
   • "챗GPT에 팁 준다고 했더니"...사소한 프롬프트로 LLM 대답 쉽게 변해
   • 구글 클라우드-허깅페이스, 전략적 파트너십 체결 "생성형 AI 및 ML 개발 가속화"
   • The retina is arguably the most impressive part of the brain.
   • llm-course - Course to get into Large Language Models (LLMs) with roadmaps and Colab notebooks.
   • 🪆Matryoshka Representation Learning (MRL) 🪆
   • Meet Jan - Bringing AI to your Desktop: Open-source ChatGPT alternative that runs 100% offline on your computer.
   • The 🤗 Hugging Face Hub now natively supports ⚡⚡ WebDataset ⚡⚡
   • AbacusAI's MetaMath-Bagel-DPO-34B sits at the top of HF🤗 LLM leader board for Maths & reasoning!
   • Matryoshka Representation Learning
   • 500개 이상의 LoRA를 만들어본 후, 터득한 비결
   • Using OpenAI text-embedding-3-large and text-embedding-3-small
   • New embedding models and API updates by OpenAI
   • AI 중심 데이터 전략에서 따져봐야 할 4가지
   • “기존 보안 툴로는 잡기 어렵다” 기업이 직면한 LLM 관련 위험 4가지
   • “환각 없이 믿을 수 있는 LLM” 검색 증강 생성(RAG)의 이해
   • 기계에게 ‘코딩’ 가르치기
   • GN⁺: Ollama, Python & JavaScript 라이브러리 공개 (ollama.ai)
   • The "Dense X Retriever" paper shows that it significantly outperforms the traditional chunk-based retriever
   • GN⁺: OpenAI, 주요 문서를 대중에게 공개하겠다는 약속을 조용히 폐기 (wired.com)
   • "생성형 AI, 2024년 전 세계 IT 지출에 큰 영향 끼치지 않는다" 가트너
   • Chatbot vs Medical Student Performance on Free-Response Clinical Reasoning Examinations
   • Most Top News Sites Block AI Bots. Right-Wing Media Welcomes Them
   • Self-Rewarding Language Models
   • RAG app running on Apple Silicon using MLX, just 3 steps
   • The Ollama Python libraries are now available:
   • InternLM-Math - 7B and 20B Chinese and English Math LMs with better than ChatGPT performances.
   • Introducing 'Prompt Engineering with Llama 2' — an interactive guide covering prompt engineering & best practices for developers, researchers & enthusiasts working with large language models.
   • AI for Tracking & Detection
   • SigLIP model pre-trained on WebLi at resolution 256x256.
   • 🚀 Checkout LLM (Large Language Models) related Code on my GitHub
   • Retrieval augmented generation (RAG) example in MLX !
   • Use RAG to build advanced text-to-SQL
   • GitHub Copilot Chat Now Generally Available
   • TensorDict is a pytorch dedicated tensor container.
   • Fooocus is an image generating software (based on Gradio).
   • 파운데이션 모델이란 무엇인가?
   • Exphormer: Scaling transformers for graph-structured data
   • Many cutting-edge computer vision models consist of multiple stages:
   • SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilities
   • Accelerating Generative AI with PyTorch IV: Seamless M4T, fast
   • Prompt Of The Year: 2023 🌟
   • 무슨 문제 많이 틀리나… 교실서 생성된 학습데이터 사교육에 넘긴다
   • Generative AI Frontline in 2023/2024
   • AI 비용을 매일 $100 에서 $1로 줄인 방법 : GPT4를 이용한 Mixtral 파인튜닝 (twitter.com/wenquai)
   • viser is a library for interactive 3D visualization + Python, inspired by tools like Pangolin, rviz, meshcat, and Gradio.
   • Non-determinism in GPT-4 is caused by Sparse MoE
   • Vertex's advanced image generation capabilities are even better when you ask Gemini to do the prompting.
   • MLX implementation of Mamba 🐍
   • Teaching CS50 with AI
   • Faster RAG re-ranking with ColBERT
   • Unlocking productivity and personalizing learning with AI
   • 법원에 AI가 뜬다…올해 9월 ‘판결문 추천 AI’ 첫 도입
   • GN⁺: WhisperSpeech – Whisper를 역으로 구축한 오픈 소스 음성합성 시스템 (github.com/collabora)
   • 모델을 재현할 수 없다면 그것은 오픈소스가 아니다 (twitter.com/amasad)
   • ActAnywhere: Subject-Aware Video Background Generation
   • QuantumReservoirPy: A Software Package for Time Series Prediction
   • Use RAG to do ArXiv Research 🌐
   • Full Stack Deep Learning - 2022 Course
   • Weaviate is now supported in DSPy 🤌
   • Sentence Embeddings. Cross-encoders and Re-ranking - Deep Dive into Cross-encoders and Re-ranking
   • Here's a MusicGen fine tune: musicgen-songstarter-v0.1 🎶
   • Vision Mamba - Efficient Visual Representation Learning with Bidirectional State Space Model
   • 🌳 Model Family Tree - Automatically calculate the family tree of a given model. It also displays the type of license each model uses (permissive, noncommercial, or unknown)
   • RAGxplorer is an interactive streamlit tool to support the building of Retrieval Augmented Generation (RAG) applications by visualizing document chunks and the queries in the embedding space.
   • Google Announces Video Generation LLM VideoPoet
   • All the Math You Missed: (But Need to Know for Graduate School)
   • LLaMA Board: A One-stop Web UI for Getting Started with LLaMA Factory
   • RAG - Vector Retrieval - A Comprehensive Study
   • DeepSpeed-FastGen: Introducing Mixtral, Phi-2, and Falcon support with major performance and feature enhancements.
   • The 10 types of clustering that all data scientists need to know. Let's dive in:
   • CoPilot for MS Office is pretty neat. Been playing with it for a bit. It's VERY tricky to sign up for though! ($20 / month). You DO NOT need an enterprise account. After a bunch of digging around this is how I enabled it on my personal/family account:
   • THE ELEMENTS OF EUCLID WITH COLOURED DIAGRAMS AND SYMBOLS
   • Self-Driving Cars — Andreas Geiger - Tübingen Machine Learning
   • AI의 근친교배
   • Stability released a 1.6b alternative to TinyLLama & Phi-2
   • Heat.js - Heat Map 시각화 라이브러리 (william-troup.com)
   • Mathematics for Deep Learning 🧑‍🎓
   • DataTrove is a library to process, filter and deduplicate text data at a very large scale.
   • nanotron - The objective of this library is to provide easy distributed primitives in order to train a variety of models efficiently using 3D parallelism.
   • Privy - 개인보호 중시 코딩 어시스턴트 (github.com/srikanth235)
   • “생성형 AI 공정 학습 여부 가린다” 페어리 트레인드, 인증 프로그램 제공
   • 호주 정부 “AI 관련 의무적 가드레일 검토 중”
   • LLM의 가장 큰 걸림돌 ‘호출 제한’을 해결하는 방법
   • ‘딥페이크, 음성 변조’… 전 세계 선거 앞두고 떠오른 오픈AI의 난제
   • 중국 군사용 AI 개발에 바이두 LLM 활용? ··· 바이두 “협력한 적 없다”
   • "두뇌를 명령이 아닌 대화 모드로 재설정하라"··· MS, 'AI 활용 습관 5가지' 제안
   • 증가하는 AI 데이터세트… 데이터 엔지니어링 및 관리에 미치는 영향
   • 387% faster TinyLlama, 6x faster GGUF conversion
   • ‘방대한 양의 과학 데이터’… NASA의 생성형 AI 도입 여정
   • 오픈AI, 군사적 용도에 대한 AI 기술 사용 약관 삭제해
   • ‘생성형 AI 도입은 원하지만 방법 아는 경영진은 부족’… 보스턴 컨설팅 조사 결과
   • Spotify's Approach to Leverage Recursive Embedding and Clustering to Enhanced Data Explainability
   • Meta is developing open source AGI, says Zuckerberg
   • Meta's AI Initiatives: $20B+ of Investment in GPUs and Data Center Infra
   • Open-source AI looks strong.
   • 5 Hard Truths About Generative AI for Technology Leaders
   • Preference Tuning LLMs with Direct Preference Optimization Methods
   • State-of-the-art Code Generation with AlphaCodium – From Prompt Engineering to Flow Engineering
   • Code Generation with AlphaCodium: From Prompt Engineering to Flow Engineering
   • VideoCrafter2: Overcoming Data Limitations for High-Quality Video Diffusion Models
   • New short course on LLMOps!
   • Overview of LLMs for Evaluation
   • Fine-grained Hallucination Detection and Editing For Language Models
   • MLX Chat - A hackish frontend for mlx-lm.
   • Tuning Language Models by Proxy
   • Introducing ASPIRE for selective prediction in LLMs
   • 🤯Inverted Whisper == Whisper Speech!
   • AIM: Autoregressive Image Models
   • Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model
   • LILO: Learning Interpretable Libraries by Compressing and Documenting Code
   • GN⁺: TinyML: 초저전력 머신러닝 기술 (ikkaro.net)
   • “GPT 스토어에 AI 여자친구 다수” 개장과 동시에 지침 위반 속출
   • AlphaGeometry: An Olympiad-level AI system for geometry
   • OpenAI Changes Its Stance, Announces That It Is Working With the Pentagon
   • WaveCoder: Widespread And Versatile Enhanced Instruction Tuning with Refined Data Generation
   • AlphaGeometry paper explanation.
   • SonicVisionLM: Playing Sound with Vision Language Models
   • VeRA: Vector-based Random Matrix Adaptation
   • AI poisoning could turn open models into destructive “sleeper agents,” says Anthropic
   • Summarizing Post Incident Reviews with GPT-4 = How we use GPT-4 to summarize incident reports.
   • The Faiss library
   • MathVista: Evaluating Math Reasoning in Visual Contexts
   • Real-time Photorealistic Dynamic Scene Representation and Rendering with 4D Gaussian Splatting
   • Democratizing LLMs: 4-bit Quantization for Optimal LLM Inference
   • Over the years we have created dozens of Computer Vision tutorials. This repository contains examples and tutorials on using SOTA computer vision models and techniques.
   • MLX Data - MLX Data is a framework agnostic data loading library brought to you by Apple machine learning research.
   • stable-code - Stable Code 3B is a model offering accurate and responsive code completion at a level on par with models such as CodeLLaMA 7B that are 2.5x larger.
   • Stable Code 3B: Coding on the Edge
   • 👉 PyTorch Image Models
   • Moore-AnimateAnyone - 캐릭터 애니메이션을 위한 이미지-to-비디오 합성 기술 (github.com/MooreThreads)
   • Query Pipeline for Advanced Text-to-SQL
   • I'm currently looking into different metrics and frameworks around Retrieval-Augmented Generation (RAG) evaluation.
   • A breakdown of the Long Context Retrieval Embedding Models from Stanford!💥
   • codeqai - Search your codebase semantically or chat with it from cli.
   • MLSys Seminars - Stanford MLSys Seminars
   • Rethinking Tabular Data Understanding with Large Language Models
   • Mix-Self-Consistency Pack - LLMs can reason over tabular data
   • PhotoMaker: Customizing Realistic Human Photos via Stacked ID Embedding
   • SD Turbo + TAESD on WebGPU with diffusers.js
   • Towards Conversational Diagnostic AI
   • Direct Preference Optimization: Your Language Model is Secretly a Reward Model
   • Listening with LLM
   • Texify - Texify is an OCR model that converts images or pdfs containing math into markdown and LaTeX that can be rendered by MathJax ($$ and $ are delimiters). It can run on CPU, GPU, or MPS.
   • Image-to-LaTeX in 3 lines of JavaScript code, with 🤗 Transformers.js!
   • Signs and Portents - Some hints about what the next year of AI looks like
   • "Machine Learning 4771" course materials from @Columbia are open.
   • 며칠 전 구글 리서치에서도 ‘LLM의 실수 식별 및 수정 가능 여부’에 대한 연구 내용을 공개했습니다.
   • LLM의 실수 식별 및 수정 가능 여부
   • A new way to search with generative AI - Get AI-powered overviews and ask follow ups, right in Search.
   • Vanna is an MIT-licensed open-source Python RAG (Retrieval-Augmented Generation) framework for SQL generation and related functionality.
   • Building a fully local LLM voice assistant to control my smart home
   • GN⁺: 스마트 홈 제어를 위한 완전 로컬 LLM 음성 비서 구축하기 (johnthenerd.com)
   • Generative AI: a game-changer that society and industry need to be ready for
   • INTERS: Unlocking the Power of Large Language Models in Search with Instruction Tuning
   • AI Voice Assistant: Enhancing Accessibility in AI with LlamaIndex and GPT3.5 (Deployed in Prod on Vercel and Render)
   • Generative AI in the Enterprise
   • Failure Points In RAG Systems
   • Deep Learning in Computer Vision with Prof. Kosta Derpanis (York University)
   • EECS 4422 Computer Vision with Prof. Kosta Derpanis
   • Sam Altman reveals in an interview with Bill Gates (2 days ago) what's coming up in GPT-4.5 (or GPT-5):
   • Retrieval meets Long Context Large Language Models
   • RAG (Retrieval augmentation generation) or longer context window LLM - what performs better? 🤔 -
   • Deep Learning for Computer Vision Michigan Online
   • (Long)LLMLingua: Enhancing Large Language Model Inference via Prompt Compression
   • The Llama Hitchiking Guide to Local LLMs
   • The AI Hierarchy of Needs
   • AMIE: A research AI system for diagnostic medical reasoning and conversations
2. 하드웨어
   • SK하이닉스 흑자 전환… ‘AI로 인한 디램 수요 증가 시사’
   • “전 세계 데이터센터 에너지 소비량…2년 내 1,000테라와트시 육박”
   • 라즈베리 파이 프로젝트에 가능성을 추가하는 8가지 확장 보드
   • 생성형 AI의 신뢰성을 높이는 3가지 기술
   • 애플 비전 프로를 사야 하는 5가지 이유
   • 이사오면서 책장을 싹 없애버렸습니다. 아이패드 이러려고 쓰는거죠? (북스캐너 캐논 R40)
   • 수명은 1만년, 전기도 필요 없다…'영화 1750편' 들어가는 유리조각
   • RAG-Survey - RAG를 잘 정리한 깃헙 문서
   • 해외 대형 출판사들, “AI 번역금지” 국내 출판사에 계약 요구
   • SCEPTER is an open-source code repository dedicated to generative training, fine-tuning, and inference, encompassing a suite of downstream tasks such as image generation, transfer, editing.
   • "2024년 XR 헤드셋 출하, 두 자릿수 성장··· 애플 비전 프로가 활성화 촉진"
   • HHKB 스튜디오 리뷰 : 토프레를 벗어난 해피해킹의 모험
   • “65% 더 많은 에너지 저장한다” 실리콘 음극재 사용한 배터리 등장
   • QA는 생성형 AI 테스트 도구를 어떻게 검증할까?
   • ‘10년 주기’ 반도체 시장 반등 가시화… “특정 제품 아닌 AI 개념이 성장 주도”
   • 허리띠 졸라매는 IT업계, "그래도 AI 서버는 사고본다"…글로벌 서버 기업 고공행진
   • CES 2024를 뒤덮은 ‘AI PC’는 소비자에게 무엇을 의미할까?
   • 삼성 갤럭시 S24 시리즈, 구글 생성 AI '제미나이 프로'와 딥마인드 AI 이미지 생성 툴 '이마젠 2' 탑재
   • Comparing the 1970’s Cray-1 supercomputer against the Raspberry Pi single-board computer range #RaspberryPi @hacksterio
   • THE VACUUM TUBE’S FORGOTTEN RIVAL
   • GN⁺: USB4를 이용하여 고속 10Gbps 풀-메쉬 네트워크 $47.98에 구축하기 (fangpenlin.com)
   • LTO(Linear Tape-Open) 드라이브
   • Running Local LLMs and VLMs on the Raspberry Pi
   • R1 - 자체 파운데이션 모델이 내장된 손바닥 크기의 디바이스 (youtube.com)
   • Struggling with Machine Learning algorithms? 🤖
   • Apple Vision Pro 코딩 시 개발자가 알아야 할 것들 (zdnet.com)
   • MediaTek to adopt Nvidia GPU in flagship mobile chip
3. 읽을거리
   • Purchasing power parities (PPP) - Total, National currency units/US dollar, 2000 – 2022
   • 대학생을 위한 실용금융(제3판)
   • 도서 출판의 미래 - 2024년과 그 이후 (linkedin.com)
   • 통수에 통수에 통수…‘든든한 아침’ 시리얼이 알고보니 막장드라마
   • 전기차 시대보다 생산 절벽의 시대가 먼저 도착한다면?
   • "주모가 차린 술·밥 먹고 공짜로 자고 가시게"...찜질방의 원조, '주막'
   • 하네다 JAL 516편 충돌 사고의 불편한 진실
   • 회사 실적에 악영항 줄 정도… 허츠, 전기차 2만 대 처분
   • 스콧 갤러웨이의 2024년 예상 (medium.com/@profgalloway)
   • 심화수학 빠지면 사교육비 줄어들까
   • Laser mapping reveals oldest Amazonian cities, built 2500 years ago

(보너스: The best math joke ever. 😂 via @mathladyhazel)

EOB