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

(오늘의 짤방: “crypto passion” leaving tech bros as they discover AI and chatGPT via @0xgaut)

1. 빅데이터/인공지능
   • Amazon CodeWhisperer - Build applications faster and more securely with your AI coding companion
   • cleanlab helps you clean data and labels by automatically detecting issues in a ML dataset.
   • Advice for ML beginners 💡 4 ways to improve your ML model:
   • The great team @MetaAI research used 2,048 80gb A100s for 5 months to train the LLaMa suite of language models
   • Sorting Data in Python With Pandas (Overview)
   • Simple Index Demo - Load documents, build the GPTSimpleVectorIndex
   • Atlassian Intelligence AI 도구 공개 (atlassian.com)
   • Here are 7 ways to access GPT-4 for free:
   • Thanks to the help of @reach_vb, try this @huggingface demo where you can chat with a PDF ! 📄🤖
   • A Very Gentle Introduction to Large Language Models without the Hype
   • Alternatives to GitHub Copilot, and some of their characteristics.
   • Double descent in human learning
   • Visual Blocks for ML: Accelerating machine learning prototyping with interactive tools
   • The evolution of the Korean particle "(으)로"
   • Can Large Language Models Transform Computational Social Science?
   • Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models
   • A simple script to get results from the OpenAI Asynchronous API
   • Improving Document Retrieval with Contextual Compression
   • Bard now helps you code
   • A simple trick to make LLMs “calibrated”
   • segment-geospatial - A Python package for segmenting geospatial data with the Segment Anything Model (SAM)
   • My Notes 📓 - This repository contains my lecture notes from graduate school on following topics 👇🏼
   • How Pytorch 2.0 Accelerates Deep Learning with Operator Fusion and CPU/GPU Code-Generation
   • The Little Book of Deep Learning
   • Inside the secret list of websites that make AI like ChatGPT sound smart
   • Automatic Prompt Optimizer for LLMs
   • GitHub Copilot Emits GPL. Codeium Does Not.
   • Stack Overflow Will Charge AI Giants for Training Data
   • 모델 서빙 최적화를 위한 프레임워크 선정과 서빙 성능 극대화하기
   • ‘챗GPT의 오픈소스 버전’··· 스테이빌리티AI, 대형 언어모델 스테이블LM 공개
   • Data analysis with SQLite and Python for PyCon 2023
   • "데이터 도움 필요하지만 데이터 때문에 괴롭다" 오라클 '기업 의사결정 딜레마' 조사
   • Whisper JAX TPU
   • Whisper JAX
   • LangChain Tutorial in Python - Crash Course
   • Announcing Google DeepMind
   • Monolith: The Recommendation System Behind TikTok
   • Bark is a transformer-based text-to-audio model created by Suno.
   • MiniChain - A tiny library for coding with large language models.
   • CompressGPT: Decrease Token Usage by ~70%
   • Awesome-Generative-RecSys - A curated list of awesome Generative Recommender Systems
   • 아마존, 분류 로봇 위한 ARM벤치 공개
   • The Anatomy of Autonomy: Why Agents are the next AI Killer App after ChatGPT
   • ChatGPT Course – Use The OpenAI API to Code 5 Projects
   • Beam - Rapidly Develop AI Projects
   • Introducing Azure OpenAI Code Repository: Your Gateway to Harnessing the Power of Generative AI
   • 프로그래밍을 위한 LLM 프롬프트 예제 (martinfowler.com)
   • AI와 LLM, 클라우드 분야의 새로운 전장
   • 오라클, ‘기업의 의사결정 딜레마’ 글로벌 조사 결과 발표… “전세계 비즈니스 리더 70%, AI에 의사결정 일임하길 원한다”
   • unyt - A package for handling numpy arrays with units.
   • GPU 비용 부담 완화책?··· “마이크로소프트 AI 연구 전용 칩 개발 중” 더인포메이션
   • Segment Anything
   • How to train your own Large Language Models
   • Stability AI Launches the First of its StableLM Suite of Language Models
   • NaturalSpeech 2: Latent Diffusion Models are Natural and Zero-Shot Speech and Singing Synthesizers
   • Create your own Custom ChatGPT
   • LangChain AI Handbook
   • Paella - 간단하고 효율적인 Text-to-Image 생성 모델 (laion.ai)
   • What is the Vercel AI Playground? - Compare and tune AI language models side-by-side:
   • 📷 📝 Multimodal C4 (mmc4) 📝 📷 - An open, billion-scale corpus of images interleaved with text.
   • 생성AI 돌풍으로 'H100' 가격 폭등...대당 최대 6000만원 호가
   • Microsoft Readies AI Chip as Machine Learning Costs Surge
   • The Complete Beginners Guide To Autonomous Agents
   • Introducing MLflow 2.3: Enhanced with Native LLM Support and New Features
   • Vocode is an open-source library for building voice-based LLM applications.
   • Understanding Large Language Models
   • Learning to Compress Prompts with Gist Tokens
   • OpenAI chief says age of giant AI models is ending; a GPU crisis could be one reason why
   • LangChain Decoded
   • DatasauRust - Blazingly fast implementation of the Datasaurus paper (500x faster than the original)
   • "문자-비디오 변환 AI 시장 연간 37.1% 성장··· 식음료 부문이 활발"
   • An example of LLM prompting for programming
   • chatGPT열풍, 공공기관의 이슈리포트를 모아보았습니다 (servicedesign.tistory.com)
   • MiniGPT-4 : 고급 LLM을 이용한 비젼-언어 이해도 향상 (minigpt-4.github.io)
   • Open Assistant - 모두를 위한 대화형 AI 공개 (open-assistant.io)
   • Web LLM - WebGPU로 브라우저에서 LLM 가속하여 실행하기 (github.com/mlc-ai)
   • A Generalist Agent
   • Endless AI templates on Replit
   • Let's build GPT: from scratch, in code, spelled out.
   • DINOv2: Learning Robust Visual Features without Supervision
   • Amazon, 생성형 AI 어플리케이션용 Bedrock 공개 (aws.amazon.com)
   • Supercharge Archive Content Discovery with ChatGPT and Azure Video Indexer
   • Simple Perplexity AI clone
   • Stanford CS330 Deep Multi-Task & Meta Learning
   • RedPajama, a project to create leading open-source models, starts by reproducing LLaMA training dataset of over 1.2 trillion tokens
   • 🦙🧪 Llama Lab 🧬🦙
   • New blog post from our CEO Prashanth: Community is the future of AI
   • Following the release of LLaMA, we saw a rapid explosion of open-source research on large language models (LLMs). Here are the three most notable model releases during this time… 🧵
   • MiniGPT-4: Enhancing Vision-language Understanding with Advanced Large Language Models
   • 프로덕션용 LLM 어플리케이션 구축하기 (huyenchip.com)
   • Auto-GPT - GPT-4를 자동화 하는 실험적 오픈소스 (github.com/Significant-Gravitas)
   • DeepSpeed Chat - RLHF를 이용한 ChatGPT-like 모델 훈련용 프레임워크 (github.com/microsoft)
   • State of the Art GPT-3 Summarizer For Any Size Document or Format
   • 한국 ‘로봇 밀도’ 세계 최고…제조업 인력난 해소 [9시 뉴스] / KBS 2023.04.16.
   • Understanding Large Language Models
   • AI 이미지가 세계적 사진 대회에서 우승한 후, 수상 거부한 아티스트
   • Google Devising Radical Search Changes to Beat Back A.I. Rivals
   • gamma - A new medium for presenting ideas. Powered by AI.
   • AgentGPT - 브라우저에 AI Agent 도입하기 (github.com/reworkd)
   • TurboPilot - 셀프호스트 가능한 Copilot 클론 (github.com/ravenscroftj)
   • The Open Assistant chat corpus just dropped, 100k-scale chat/instruction dataset from thousands of participants.
   • Building LLM applications for production
   • BEN'S BITES - a daily feed of AI product launches and news
   • A simple vector database for 100K vectors with the 1536 dim OpenAI embeddings.
   • 우울한 AI 연구자를 위한 생존전략 2023.3
   • Prompt Engineering vs. Blind Prompting
   • Web LLM
   • kNN vs. SVM
   • Why is 𝗠𝗼𝗱𝗲𝗹 𝗥𝗲𝗴𝗶𝘀𝘁𝗿𝘆 so important in your 𝗠𝗟𝗢𝗽𝘀 𝗦𝘁𝗮𝗰𝗸?
   • Matrix Calculus
   • What Are Transformer Models and How Do They Work?
   • Graph classification with Transformers
   • 🍣🔗 BentoChain A 🦜🔗 LangChain deployment example using 🍱 BentoML.
   • Prompt injection: What’s the worst that can happen?
   • Monolith: The Recommendation System Behind TikTok
   • API로 AI 모델 맞춤 활용··· AWS, 기업용 생성형 AI 모델 서비스 ‘아마존 베드록’ 공개
   • 10 Graphs That Sum Up the State of AI in 2023
   • 구글 데이터플렉스 리뷰 | ‘프리뷰 단계인데도’ 데이터 사일로의 완전한 대안
   • Grounded Segment Anything: From Objects to Parts
   • Grounded Segment Anything：From Objects to Parts
   • AGIEval - This repository contains information about AGIEval, data, code and output of baseline systems for the benchmark.
   • The OpenAI Cookbook shares example code for accomplishing common tasks with the OpenAI API.
   • databricks-dolly-15k is an open source dataset of instruction-following records used in training databricks/dolly-v2-12b that was generated by thousands of Databricks employees in several of the behavioral categories outlined in the InstructGPT paper, including brainstorming, classification, closed QA, generation, information extraction, open QA, and summarization.
   • One Small Step for Generative AI, One Giant Leap for AGI: A Complete Survey on ChatGPT in AIGC Era
   • 91% of ML Models Degrade in Time
   • elevenlabs - Prime Voice AI
   • AI 특이점은 이미 도래했다
   • Academic writing is hard. Writefull’s AI helps you write, paraphrase, copyedit, and more.
   • Segment Anything Model and the Hard Problems of Computer Vision — with Joseph Nelson of Roboflow
   • Auto-GPT Unmasked: The Hype and Hard Truths of Its Production Pitfalls
   • The ‘Godfather of AI’ Says Doomsayers Are Wrong and ChatGPT Isn’t All That Innovative
   • ChatGPT와 텔레그램 및 음성을 이용한 컨시어지 봇 오픈소스 (github.com/RafalWilinski)
   • Announcing New Tools for Building with Generative AI on AWS
   • GPT Unicorn: A Daily Exploration of GPT-4's Image Generation Capabilities
   • What Kind of Mind Does ChatGPT Have?
   • Go smol or go home - Why we should train smaller LLMs on more tokens
   • 쿠팡이 '기계학습'으로 물류 입고 프로세스 개선한 방법
   • AI is already taking video game illustrators’ jobs in China
   • ChatGPT is better at predicting how stocks will react to news headlines than traditional models, new study shows
   • LangChain - LLM을 외부와 연결해주는 라이브러리 (github.com/hwchase17)
   • ChatGPT Beyond English: Towards a Comprehensive Evaluation of Large Language Models in Multilingual Learning
   • Do you actually need a vector database?
   • Building LLM applications for production
   • 💧 A Watermark for Large Language Models
   • How does Stable Attribution work?
   • GLAZE: Protecting Artists from Style Mimicry by Text-to-Image Models
   • Coding Sucks Anyway — Matt Welsh on the End of Programming
   • 미드저니 Zarya of the Dawn 소송건 판결
   • Tabby - 셀프호스트 가능한 오픈소스 AI 코딩 비서 (github.com/TabbyML)
   • 마이크로소프트, 엣지 브라우저에 ‘AI 이미지 생성기’ 통합
   • Midjourney AI Guide
   • TRL - Transformer Reinforcement Learning
   • OpenAI, 1단계 만으로 생성 가능한 Consistency Model의 코드 공개 (github.com/openai)
   • Media Issue 9권 3호 <챗GPT 이용 경험 및 인식 조사>
   • Free Dolly: Introducing the World's First Truly Open Instruction-Tuned LLM
   • Boosted Prompt Ensembles for Large Language Models
   • phind - 개발자를 위한 GPT-4 기반 검색 엔진 (phind.com)
   • DeepSpeed Chat: Easy, Fast and Affordable RLHF Training of ChatGPT-like Models at All Scales
   • Chat with your favourite LLaMA models - LlamaChat allows you to chat with LLaMa, Alpaca and GPT4All models1 all running locally on your Mac.
   • Teaching Large Language Models to Self-Debug
   • Edit & Generate Anything by Segment-Anything (github.com/sail-sg)
   • Machine Learning Model Deployment — A Simple Checklist
   • go-openai: OpenAI ChatGPT, GPT-3, GPT-4, DALL·E, Whisper API wrapper for Go
   • How does Twitter recommend “For You” Timeline in 1.5 seconds?
   • TurboPilot is a self-hosted copilot clone which uses the library behind llama.cpp to run the 6 Billion Parameter Salesforce Codegen model in 4GiB of RAM.
   • Why ChatGPT and Bing Chat are so good at making things up
   • gpt4-tokenizer
   • Diffusion Models for Medical Image Analysis: A Comprehensive Survey
   • One of the main benefits of GPT-4 relative to prior models (like ChatGPT/GPT-3.5) is that the model is incredibly steerable.
   • MLOps Zoomcamp
   • Cerebras-GPT: A Family of Open, Compute-efficient, Large Language Models
   • A List of 1 Billion+ Parameter LLMs
   • The foundational model market is already fragmented. There are over 50 one billion+ parameter LLMs to choose from (open-source or proprietary API).
   • Zero-Shot Next-Item Recommendation using Large Pretrained Language Models
   • Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matters
   • OpenAGI: When LLM Meets Domain Experts
   • Prompt Reducer - GPT-4 프롬프트 압축하기 (promptreducer.com)
   • openplayground - An LLM playground you can run on your laptop.
   • 5 latest open-source LLMs (save the list)
   • 🚨 Dangermode is a ChatGPT Plugin written with Python and FastAPI that allows ChatGPT to execute code snippets in an IPython session, whether it's the console, the notebook, or a JupyterLab session.
   • Basic Molecular Representation for Machine Learning
   • A Recipe for Training Large Models
   • Why is GPT-3 15.77x more expensive for certain languages?
   • 1/ Memory limitations in GPT can lead to loss of context in long conversations. Here are four techniques to effectively manage conversations and get meaningful responses. 👇
   • langchain-pdf-qa - This repo lets you use a local PDF/text file to ask questions and generate asnwers.
   • Grounded-Segment-Anything - We plan to create a very interesting demo by combining Grounding DINO and Segment Anything which aims to detect and segment Anything with text inputs!
   • Enhancing ChatGPT With Infinite External Memory Using Vector Database and ChatGPT Retrieval Plugin
   • @gpt_index now has a BIG feature to help with temporal data retrieval + LLM’s: Recency Filtering / Filtering outdated Nodes ⌛️
   • Stanford benchmarks and compares numerous Large Language Models
   • Integrating ChatGPT with internal knowledge base and question-answer platform
   • Analysis of Twitter the-algorithm source code with LangChain, GPT4 and Deep Lake
   • 1/ 🧵So how do we overcome 4096 token limit in OpenAI GPT requests?
   • Recommendation Alignment (RecAlign)
   • Understanding Deep Learning(PDF)
   • In recent study by MSFT, few LLMs were evaluated and ranked based on performance across multiple tasks. The findings are as follows:
   • Formal Algorithms for Transformers
   • 논문 보다가 아예 모르는 단어 나와서 설명 요청할 때 이 프롬프트 괜찮은 것 같다. What is {keyword}? Explain step by step to a novice, and used examples if possible.
   • 생성 AI 앱이 B2B 시장에서 통하기 위한 조건
   • From Deep to Long Learning?
   • How to Divide by Zero
   • Stability AI is on shaky ground as it burns through cash and looks at a management overhaul
   • Faust - Programming Language for Audio Applications and Plugins
   • Eight Things to Know about Large Language Models
   • MIT Introduction to Deep Learning
   • Instruction Tuning with GPT-4
   • llm-chain is a collection of Rust crates designed to help you work with Large Language Models (LLMs) more effectively.
   • Sparks of Artificial General Intelligence: Early experiments with GPT-4
   • AGI 시대: 다음 세대의 스타트업
   • AX/Diffusers community sprint 🧨
   • HuggingGPT - A system to connect LLMs with ML community.
   • Hacking Google reCAPTCHA v3 using Reinforcement Learning
2. 하드웨어
   • New 3D Printable Search Engine
   • 자동 완성 ① 한자 폐기론
   • Booting modern Intel CPUs
   • RISC-V GNU Compiler Toolchain
   • “히타치 HDD 수명 가장 짧아…맥스터가 내구성 우수” 시큐어 데이터 리커버리
   • Sony backs maker of tiny Raspberry Pi computers with fresh funding, access to A.I. chips
   • Sega 3-D Glasses: How did they work?
3. 블록체인과 메타버스
   • 이창용 "초고속 뱅크런 대비 필요, 지급보증 담보자산 상향 검토"
   • ‘264억 손실’ 에듀테크에서 손 떼는 이투스에듀… “학원 사업에 집중”
   • California DMV puts car titles on Tezos blockchain, consumer applications on the way
   • Steps towards prompt-based creation of virtual worlds
   • 메타, 메타버스 대신 인공지능에 집중 - 경영진 인공지능에 대부분 시간 할애…"연내 기술 활용한 앱 선보일 것"
4. 읽을거리
   • Study finds new pathway for clearing misfolded proteins
   • Space Elevator
   • MISC50 Cognitive Biases in the Modern World
   • KB부동산시장 리뷰 2023-4호
   • Meta Tries to Lure Advertisers With Reels Discounts, AI Tools
   • 수수께끼 같은 미국경제: 이렇게 이해하자
   • 2018년 국민여행조사-해외여행 경험 22%, 인터넷보다 지인 신뢰, 연휴보다 평일 출발
   • Thought Examinations, Indoctrination Meetings and Struggle Sessions
   • 상사와 불화로 퇴사, 청년 모차르트의 ‘해방 일지’
   • 사라지는 ‘대졸 임금 프리미엄’… 미·영, 위기의 대학 시장
   • 대학 서열은 돈의 서열이다
   • '1000억 자산' 세이노 "부모 탓, 자기 삶 파괴…모녀 슬픈 눈동자, 내 거울"
   • 넷플릭스만 살아남는 OTT 시장? 빨간불 켜진 지속가능성
   • How Killing Sparrows Led to Great Famines in China
   • 흔들리는 달러 패권 美, 희생양 찾고있다
   • Ever-expanding animation of the life of the 796th floor of a space station

보너스: 💡 An overview of the workflow in machine learning engineering! via @DataScienceDojo

EOB