[B급 프로그래머] 4월 3주 소식(빅데이터/인공지능, 하드웨어, 읽을거리 부문)

(오늘의 짤방: 주요 인공지능 모델들의 트레이닝 비용. 잼민이 트레이닝 시키는데에만 $191M (2600억원) 정도 들었다고 예상함. via @mynameisdjkim)

1. 빅데이터/인공지능
   • 슬랙 "생성형 AI 추가 요금은 인당 월 10달러"
   • 메타, 최신 LLM ‘라마3’ 오픈소스 공개…“기술ㆍ코딩ㆍ역사 등 다양한 데이터 학습”
   • Meta, Llama 3 공개 (ai.meta.com)
   • Mark Zuckerberg - Llama 3, $10B Models, Caesar Augustus, & 1 GW Datacenters
   • Meta Llama 3에 대한 Andrej Karpathy의 평가 (twitter.com/karpathy)
   • 메타, 페북 등에 AI 챗봇 '메타 AI' 탑재…최신 라마3 기반
   • Llama-3 tokenizer playground
   • Pegasus-1 Technical Report by 트웰브랩스
   • We are releasing two new open-source reranker models: jina-reranker-v1-turbo-en and jina-reranker-v1-tiny-en, the latter has only 30M parameters and four layers! 🤯 These two new rerankers enjoy 5X faster inference speed than our last jina-reranker-v1-base model at only a very small cost on the quality.
   • More technical details on the new Meta Llama 3 models announced today. 🦙🧵
   • Imagine Flash: Accelerating Emu Diffusion Models with Backward Distillation
   • Meet Your New Assistant: Meta AI, Built With Llama 3
   • Vision language models, explained 🧶❤️
   • Build a RAG system with Llama 3B-Instruct for your PDFs
   • From Image to UML: First Results of Image Based UML Diagram Generation Using LLMs
   • #LLaMA3 is out! It's the same architecture as Llama-2, except for some differences:
   • LLaMA-3 is a prime example of why training a good LLM is almost entirely about data quality…
   • Here's all that we know about Meta Llama 3 so far
   • 'AI 사각지대 이해하기'… 하이브리드 환경 전반에서 AI를 식별하고 보호하려면
   • "LLM 시장, 2030년까지 연평균 33.2% 성장" 마켓앤마켓 보고서
   • 이직할 때 ‘우리 회사 데이터 분석 잘한다’는 말에 낚이지 않는 방법
   • The Assistants API is designed to help developers build powerful AI assistants capable of performing a variety of tasks.
   • '세금 결산을 지원하는 생성형 AI 환경 구축'… H&R블록 CIO
   • 한국어 영향 미칠까?··· 오픈AI, 일본 지사 설립 및 일본어 맞춤 GPT-4 모델 공개
   • Stable Diffusion 3 API 이용 가능 (stability.ai)
   • LG CNS, ‘DAP 젠AI 플랫폼’ 고도화…“생성형 AI 4대 솔루션 모두 제공”
   • “프레임 추가하고 배우 의상까지 바꾼다” 어도비, 프리미어 프로에 AI 신기능 추가
   • GN⁺: AI에 호기심 있는 앱 개발자를 위한 좋은 시작점, Embeddings (bawolf.substack.com)
   • SOAR: New algorithms for even faster vector search with ScaNN
   • GN⁺: Mistral AI, 새로운 오픈 모델 Mixtral 8x22B 공개 (mistral.ai)
   • torchtune: Easily fine-tune LLMs using PyTorch
   • Video2Game: Real-time, Interactive, Realistic and Browser-Compatible Environment from a Single Video
   • Code with CodeQwen1.5
   • 👉 PixArt-Σ: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generation(데모)
   • Magic Clothing is a branch version of OOTDiffusion, focusing on controllable garment-driven image synthesis
   • The following examples show how to get started with the unstructured library.
   • LLM-Powered Test Case Generation for Detecting Tricky Bugs
   • HairFastGAN: Realistic and Robust Hair Transfer with a Fast Encoder-Based Approach
   • Current best local model:
   • 스탠포드의 2024년 AI 현황을 설명하는 15가지 그래프 (spectrum.ieee.org)
   • 구글 딥마인드의 ALOHA Unleashed - 저비용 로봇과 AI로 손재주 한계를 뛰어넘는 중 (twitter.com/tonyzzhao)
   • Vision Transformers 비쥬얼 가이드 (blog.mdturp.ch)
   • Leave No Context Behind: Efficient Infinite Context Transformers with Infini-attention
   • 오라클 DB의 아버지, 앤디 멘델손… “굳이 LLM 만들 필요 없어”
   • MMInA: Benchmarking Multihop Multimodal Internet Agents
   • Extending Llama to a new language
   • 어도비, 동영상 편집기 '프리미어'에 '소라' 등 결합 예정..."획기적 통합"
   • 검색에 잘 걸리게 하는 SEO(search engine optimization)에서 이제는 LLM의 추천을 잘 받는 LSO(LLM-based search optimization)의 시대가 온다.
   • Colab 101: Your Ultimate Beginner's Guide!
   • Compression Represents Intelligence Linearly
   • 'AI 헤드셋 착용한 직원'… 리테일 매장을 AI로 혁신 중인 트랙터 서플라이
   • “짝퉁 식별부터 AI 패션쇼까지” 명품 시장에 부는 ‘생성형 AI’ 바람
   • ACCELERATING SCIENTIFIC DISCOVERY WITH GENERATIVE KNOWLEDGE EXTRACTION, GRAPH-BASED REPRESENTATION, AND MULTIMODAL INTELLIGENT GRAPH REASONING
   • Pile-T5 - Trained T5 on the Pile
   • Introducing Idefics2: A Powerful 8B Vision-Language Model for the community
   • Build a Large Language Model (From Scratch) source code
   • OpenAI, 일본 지사 설립후 일본어 최적화 GPT-4 모델 출시 (openai.com)
   • Welcome to the Gemini API Cookbook
   • We’re excited to release a brand-new tutorial series 💫 : an introduction to agents and tools 🎬🤖
   • 당근에서 LLM 활용하기
   • “AI가 변환한 흑백 사진·영화 멋진데”... GIST, 사진·영상을 전문가 수준으로 컬러에서 흑백 보정하는 AI 기술 개발
   • LLaMA Now Goes Faster on CPUs
   • Introducing OpenAI Japan
   • Scaling (Down) CLIP: A Comprehensive Analysis of Data, Architecture, and Training Strategies
   • "문장 입력하면 영상 뚝딱"…구글·어도비 동영상 생성 AI 참전
   • CS 886: 파운데이션 모델의 최신 연구 동향 (cs.uwaterloo.ca)
   • Using Python for Data Analysis
   • Building LLMs for Code Repair by replit
   • Can GPT Optimize My Taxes?
   • ⚡ LitGPT - Pretrain, finetune, evaluate, and deploy 20+ LLMs on your own data
   • Open-Sora: Democratizing Efficient Video Production for All
   • AIChat - All-in-one chat and copilot CLI that integrates 10+ AI platforms.
   • MiniCPM-V - MiniCPM-V 2.0: An Efficient End-side MLLM with Strong OCR and Understanding Capabilities
   • The Anatomy of Autonomy: Why Agents are the next AI Killer App after ChatGPT
   • Gemini Agent Example
   • Grok-1.5 Vision Preview: 고급 시각 처리 기능을 갖춘 최초의 멀티모달 모델 소개 (x.ai)
   • Adobe's Firefly Image Generator Was Partially Trained on AI Images From Midjourney, Other Rivals
   • Building a better repository map with tree sitter
   • Show GN: AI 한영/영한 번역기 도다리 (github.com/vEduardovich)
   • AI를 활용하는 데 도움이 되는 챗GPT 프롬프트 프레임워크
   • Rho-1: Not All Tokens Are What You Need by Microsoft
   • 📘 LangChain 한국어 튜토리얼
   • 멀티모달 검색 증강 생성 101
   • Linear Regression is one of the most important tools in a Data Scientist's toolbox. Here's everything you need to know in 3 minutes.
   • 생성 AI를 통해 모델 이미지 문제 해결하기 - AI를 활용한 이미지 생성부터 문제 해결까지
   • Deep-Learning-in-Production - In this repository, I will share some useful notes and references about deploying deep learning-based models in production.
   • “복잡한 쿼리 작성 AI에 맡기세요”··· 스노우플레이크, AI 기반 SQL 어시스턴트 출시
   • MusicGen is a text-to-music model capable of genreating high-quality music samples conditioned on text descriptions or audio prompts.
   • Introducing MusicGen Web: AI-powered music generation directly in your browser, built with 🤗 Transformers.js! 🎵
   • LLMs Can Do Regression(참고: Game Over for traditional ML methods)
   • Dot - 로컬 LLM과 RAG을 쉽게 사용하게 해주는 스탠드얼론 오픈소스 앱 (github.com/alexpinel)
   • LaVague is an open-source project designed to automate automation for devs!
   • CLIcK: A Benchmark Dataset of Cultural and Linguistic Intelligence in Korean
   • 단계별로 알아보는 완전 로컬 검색 증강 생성(RAG)
   • GPTScript is a new scripting language to automate your interaction with a Large Language Model (LLM), namely OpenAI.
   • The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits
   • ToxicChat: Unveiling Hidden Challenges of Toxicity Detection in Real-World User-AI Conversation
   • I used generative AI to turn my story into a comic—and you can too
   • Lore Machine is a story visualization systems
   • Character Consistency in Stable Diffusion
   • Prompt injection and jailbreaking are not the same thing
   • Researchers jailbreak AI chatbots with ASCII art -- ArtPrompt bypasses safety measures to unlock malicious queries
   • ASCII art elicits harmful responses from 5 major AI chatbots
   • You can now train a 70b language model at home(using 2 GPUs)(FSDP and QLoRA)
   • LLM4Decompile - Reverse Engineering: Decompiling Binary Code with Large Language Models
   • LLM4Decompile: Decompiling Binary Code with Large Language Models
   • Nvidia’s CEO, Jensen Huang: AI will take over coding, making learning optional
   • Nvidia announces “moonshot” to create embodied human-level AI in robot form
   • Machine Learning University (MLU) is an education initiative from Amazon designed to teach machine learning theory and practical application.
   • GPT-4 Turbo with Vision is a step backwards for coding
   • GN⁺: Aider - 터미널에서 AI와 페어 프로그래밍 하기 (github.com/paul-gauthier)
   • This repository contains a lightweight library for evaluating language models. We are open sourcing it so we can be transparent about the accuracy numbers we're publishing alongside our latest models (starting with gpt-4-turbo-2024-04-09).
   • Chronon - AirBnb가 공개한 오픈소스 ML 피쳐 플랫폼 (github.com/airbnb)
   • Jan - Rethink the Computer. Turn your computer into an AI machine
   • The tech industry can’t agree on what open-source AI means. That’s a problem.
   • [독서광] 쉽고 빠르게 익히는 실전 LLM(본인이 작성한 글)
   • narwhals - Extremely lightweight and extensible compatibility layer between Polars, pandas, Modin, and cuDF (and more!).
   • “앱 구축하는 것보다 어렵다” LLM을 테스트하는 4가지 전략
   • AI의 발전, LLM을 넘어서는 사고가 필요하다
   • “AI에 효과적으로 말 걸기” 똑똑한 프롬프트 엔지니어링의 10가지 비밀
   • In this paper, we introduce TimeGPT, the first foundation model for time series, capable of generating accurate predictions for diverse datasets not seen during training.
   • OpenAI, GPT-4 Turbo with Vision 일반 공개(GA) (platform.openai.com)
   • Open-Parse - LLM을 위한 PDF 레이아웃 분할(Chunking) (github.com/Filimoa)
   • UBS "AI 랠리 지속"…올해 기술주 이익 18% 급증 전망
   • 올거나이즈, ‘금융 전문 LLM 리더보드’ 공개
   • Multilingual Large Language Model: A Survey of Resources, Taxonomy and Frontiers
   • Anthropic, Claude에 함수 호출(Tool Use) 기능 베타 공개 (docs.anthropic.com)
   • Distributed tuning with Gemma using Keras
   • LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code
   • GLiNER is a Named Entity Recognition (NER) model capable of identifying any entity type using a bidirectional transformer encoder (BERT-like).
   • LG유플러스도 AI챗봇 경쟁 합류…6월부터 자체 LLM 적용
   • 운전자 습관·행동 읽어 `척척`… 티맵, AI로 진화한다
   • Gemini 1.5 Pro Now Available in 180+ Countries; With Native Audio Understanding, System Instructions, JSON Mode and More
   • March RAGness - We pitted the top RAG systems against each other and used Tonic Validate to rank their performance.
   • 오픈AI-메타, 차세대 LLM 곧 출시...추론 역량 중무장 예고
   • 비자, 플레이드, 네트워크, 과업 (번역)
   • MiniGPT4-Video: Advancing Multimodal LLMs for Video Understanding with Interleaved Visual-Textual Tokens
   • GN⁺: 진정한 개방형 LLM, Hello OLMo 등장 (blog.allenai.org)
   • CodeGemma - an official Google release for code LLMs
   • JP모건 회장 "AI, 증기기관 발명에 비견…산업 지형 바꿀 것"
   • Text Generation Inference (TGI) is a toolkit for deploying and serving Large Language Models (LLMs).
   • SwapAnything: Enabling Arbitrary Object Swapping in Personalized Visual Editing
   • Navigating LLM Deployment: Tips, Tricks and Techniques by Meryem Arik at QCon London
   • MA-LMM: Memory-Augmented Large Multimodal Model for Long-Term Video Understanding
   • llm.c - raw C/CUDA 로 LLM 훈련하기 (github.com/karpathy)
   • llm.c - LLM training in simple, pure C/CUDA. There is no need for 245MB of PyTorch or 107MB of cPython.
   • ML for Developers - Learn how to combine machine learning with software engineering to design, develop, deploy and iterate on production ML applications.
   • GN⁺: Meta, 다음 주에 Llama 3의 소형 버전 출시 예정 (theinformation.com)
   • lumentis - AI로 녹취록과 비정형 정보에서 자동 문서 생성 (github.com/hrishioa)
   • Ollama supports embedding models, making it possible to build retrieval augmented generation (RAG) applications that combine text prompts with existing documents or other data.
   • OneLLM: One Framework to Align All Modalities with Language. 📌 Through a unified multimodal encoder and progressive alignment pipeline, OneLLM connects vision, audio, video, and more to language seamlessly.
   • Using LlamaParse for Knowledge Graph Creation from Documents
   • AI-Powered Search: Embedding-Based Retrieval and Retrieval-Augmented Generation (RAG)
   • Ferret-UI: Grounded Mobile UI Understanding with Multimodal LLMs
   • LLM-Based Multi-Agent Systems for Software Engineering: Vision and the Road Ahead
   • 5 GitHub repositories that will give you superpowers as an AI/ML Engineer:
   • GN⁺: AI에 패한 후 프로 바둑 기사들, 더 뛰어나고 창의적으로 변화 (henrikkarlsson.xyz)
   • 2024년 ML/AI/Data 업계 지도와 최신 트렌드 (mattturck.com)
   • Loki by LibrAI - 사실 여부 검증 프로세스를 자동화하는 오픈소스 도구 (github.com/Libr-AI)
   • Running OCR against PDFs and images directly in your browser
   • 🤖 Machine Learning Tutorials Repository 🤖
   • InstantStyle: Free Lunch towards Style-Preserving in Text-to-Image Generation
   • More Agents Is All You Need
   • Open-Sora-Plan - We are thrilled to present Open-Sora-Plan v1.0.0, which significantly enhances video generation quality and text control capabilities.
   • FreeAskInternet is a completely free, private and locally running search aggregator & answer generate using LLM, Without GPU needed.
   • In this tutorial, we'll see how to use LlamaIndex Instrumentation module to send intermediate steps in a RAG pipeline to the frontend for an intuitive user experience.
   • GN⁺: Cohere의 Command R+ - 비즈니스를 위해 구축된 확장 가능한 LLM (txt.cohere.com)
   • SWE-bench - Can Language Models Resolve Real-World GitHub Issues?
   • Opera 브라우저, 사용자가 LLM을 다운받아 로컬로 실행하는 기능 추가 (techcrunch.com)
   • JetMoE: Reaching LLaMA2 Performance with 0.1M Dollars
   • LLM Visualization
   • StarCoder2 is a family of code generation models (3B, 7B, and 15B), trained on 600+ programming languages from The Stack v2 and some natural language text such as Wikipedia, Arxiv, and GitHub issues.
   • The Illustrated Stable Diffusion
   • Praison AI, leveraging both AutoGen and CrewAI or any other agent framework, represents a low-code, centralised framework designed to simplify the creation and orchestration of multi-agent systems for various LLM applications, emphasizing ease of use, customization, and human-agent interaction.
   • Engineering household robots to have a little common sense
   • Build an LLM RAG Chatbot With LangChain
   • Evalverse - The Universe of Evaluation. All about the evaluation for LLMs. by Upstage
   • How to Combine Vector Search and SQL for RAG
   • Scaling AI/ML Infrastructure at Uber
   • RAG 애플리케이션을 파일럿 단계에서 프로덕션 단계로 전환하는 4가지 방법
2. 하드웨어
   • 우리집 '전기도둑'은 누굴까? 가전제품별 소비전력 알아보기
   • 다시 뜨거운 데스크톱 CPU 시장…'윈텔'의 아성은 언제까지 공고할 것인가
   • 인텔 ‘AI 에브리웨어 전략’, 엔비디아의 시장 독점 허물까
   • GN⁺: AMD, Ryzen Pro 8000 시리즈 프로세서 공개 (tomshardware.com)
   • Vortex - OpenCL 호환 가능한 RISC-V 아키텍처 기반의 풀스택 오픈소스 GPGPU (vortex.cc.gatech.edu)
   • 메타, ‘차세대 MTIA’ 발표··· "1세대 AI 칩 대비 3배 성능"
   • “인텔의 최근 혁신은 솔직히 지루했다"··· 겔싱어, 인텔 AI 로드맵 발표
   • 로봇 청소기는 어떻게 길을 찾고 장애물을 피할까
   • '가우디 기반 생태계 구축'… 네이버클라우드-인텔, AI 공동연구센터 설립
   • 다이슨, 청소 궤적을 실시간 시각화하는 AR 도구 다이슨 클린트레이스™ 공개
   • Meta, 차세대 학습 및 추론 가속기 MTIAv2 공개 (ai.meta.com)
   • GN⁺: 인텔 Gaudi 3 AI 가속기 출시 (intel.com)
3. 읽을거리
   • 성인 10명중 6명, 1년에 책 한 권도 안 읽는다…독서율 최저치 경신
   • A chunk of metal that tore through a Florida home definitely came from the ISS
   • [단독] ‘메타버스 서울’ 혈세 60억원 날렸다
   • '세계 최초' 기술이라더니…한국 연구진 조작 논란에 '발칵'
   • 매달 15만원 내면 '월 60만원' 준다더니…日 기초연금 결국
   • 2024 보통사람 금융생활 보고서 - 신한은행
   • Korean Won Topped Dollar as Preferred Currency for Crypto Trades in First Quarter
   • Can you build a Biden/Trump voter?
   • The cheaper the brand, the more polyester they use
   • 신선식품은 어쩌다 빈부격차의 상징 됐나 [마켓톡톡]
   • 12년째 이어진 수학계 '초전도체급
   • 맹모'삼천만원'지교 시대... 자녀교육과 노후준비 균형 맞추려면
   • 돈이 얼마나 있어야 생각없이 소비할 수 있을까?
   • 조중동 매출이 한겨레‧경향 5배.
   • "그냥 멍청히 서 있어요" 통화주의 반격…이래야 경제 돌아간다?
   • 60세 이상 감독은 10% 감점, 전작 망했으면 가점… 수익률 176%의 비결
   • [친절한 경제] 고물가·고금리 당분간 더 간다…멀어지는 금리인하 기대
   • What "Made in China" Means Today
   • 여의도 전업 트레이더 썰.ssul
   • Audio Dialogues: Dialogues dataset for audio and music understanding
   • 설마했는데…“이건 쇼크다” 발칵 뒤집힌 ‘국민 메신저’

(보너스: Expectation vs Reality. via @levikul09)

EOB