Eric Chang
26 followers · 22 following · 731 views
on the atlas — 94
- Voyager | An Open-Ended Embodied Agent with Large Language Models4 savers
- [2206.11795] Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos2 savers
- dall-e-2.pdf2 savers
- Netflix3 savers
- Home | Helicone - Monitoring for Generative AI2 savers
- RLHF: Reinforcement Learning from Human Feedback9 savers
- _______________🚗_____8 savers
- Transformers Explained Visually (Part 1): Overview of Functionality | by Ketan Doshi | Towards Data Science2 savers
- Prompt Engineering Guide | Prompt Engineering Guide7 savers
- LLM Evaluation doesn't need to be complicated2 savers
- The GraphRAG Manifesto: Adding Knowledge to GenAI2 savers
- GraphRAG: New tool for complex data discovery now on GitHub - Microsoft Research1 savers
- Rewind6 savers
- GPT-4o doesn't consistently respect JSON schema on tool use - API / Bugs - OpenAI Developer Forum1 savers
- ADPList: Learn from the world's best mentors for free2 savers
- The Design Sprint — GV2 savers
- Lindy.ai — Meet Your AI Employee7 savers
- cjwbw/seamless_communication – Run with an API on Replicate1 savers
- Perplexity AI: revolutionizing information search and discovery | by Fabio Lalli | IQUII | Dec, 2023 | Medium1 savers
- open-compass/opencompass: OpenCompass is an LLM evaluation platform, supporting a wide range of models (LLaMA, LLaMa2, ChatGLM2, ChatGPT, Claude, etc) over 50+ datasets.2 savers
- Introduction1 savers
- Cursor - The AI-first Code Editor4 savers
- deepset-ai/haystack: Haystack is an open source NLP framework that leverages pre-trained Transformer models. It enables developers to quickly implement production-ready semantic search, question answering, summarization and document ranking for a wide range of NLP applications.2 savers
- JSON Formatter & Validator2 savers
- Why are AI Products Doomed to Fail? | by Jeremy Arancio | Nov, 2023 | Towards AI3 savers
- GraphCast: AI model for faster and more accurate global weather forecasting - Google DeepMind3 savers
- 🦜️🏓 LangServe | 🦜️🔗 Langchain1 savers
- jason (@jxnlco) / X1 savers
- LongLLMLingua: Bye-bye to Middle Loss and Save on Your RAG Costs via Prompt Compression | by Huiqiang Jiang | Nov, 2023 | LlamaIndex Blog1 savers
- Stack AI: The Middle-Layer of AI1 savers
- Unstructured1 savers
- n8n LangChain Integration2 savers
- Mukosame/Anime2Sketch: A sketch extractor for anime/illustration.1 savers
- Knowledge management and documentation for code | Swimm1 savers
- Beyond Text: Making GenAI Applications Accessible to All1 savers
- Advanced Topics in Natural Language Processing2 savers
- jxnl/instructor: openai function calls for humans1 savers
- [2305.14251] FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation1 savers
- [2309.11495] Chain-of-Verification Reduces Hallucination in Large Language Models1 savers
- Andrei Kovalev's Midlibrary: Midjourney AI Styles Library and Guide1 savers
- LangChain ParentDocumentRetriever: Strike a Balance between large vs small chunks | ClusteredBytes1 savers
- LLM Evaluation Metrics2 savers
- https://memprompt.com2 savers
- OpenAI4 savers
- DALL·E2 savers
- Seq2seq and Attention2 savers
- Google AI Blog: Exploring Transfer Learning with T5: the Text-To-Text Transfer Transformer2 savers
- A successful Git branching model » nvie.com5 savers
- Can we intentionally improve the world? Planners vs. Hayekians – Julia Galef5 savers
- Why I do what I do | Jocelyne Murphy58 savers
- What's going on here, with this human? - Graham Duncan Blog49 savers
- Fast · Patrick Collison48 savers
- Learn In Public30 savers
- The Technium: 103 Bits of Advice I Wish I Had Known28 savers
- Finding Person-Problem Fit - Compound Manual27 savers
- transformer_attention.pdf23 savers
- Excalidraw | Hand-drawn look & feel • Collaborative • Secure20 savers
- [2304.03442] Generative Agents: Interactive Simulacra of Human Behavior19 savers
- How to Pick Your Life Partner - Part 2 — Wait But Why18 savers
- Raycast16 savers
- Perplexity AI: Ask Anything16 savers
- LLM Powered Autonomous Agents | Lil'Log14 savers
- Illustrating Reinforcement Learning from Human Feedback (RLHF)13 savers
- Attention Traps - by Varun - Public Experiments12 savers
- thesephist.com12 savers
- What every computer science major should know11 savers
- DALL·E 211 savers
- [2304.03442] Generative Agents: Interactive Simulacra of Human Behavior10 savers
- The Arc of the Practical Creator9 savers
- Logseq: A privacy-first, open-source knowledge base8 savers
- containment8 savers
- Elicit | The AI Research Assistant7 savers
- The data model behind Notion's flexibility7 savers
- Mercury7 savers
- The Illustrated Stable Diffusion – Jay Alammar – Visualizing machine learning one concept at a time.6 savers
- [2005.14165] Language Models are Few-Shot Learners6 savers
- OpenAI's "Planning For AGI And Beyond" - by Scott Alexander5 savers
- Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges5 savers
- [2201.11903] Chain of Thought Prompting Elicits Reasoning in Large Language Models5 savers
- Gandalf | Lakera - Prompt Injection4 savers
- Monocle4 savers
- The Next Chapter of Readwise: Our Own Reading App4 savers
- Stop talking about AI ethics. It’s time to talk about power.4 savers
- Eleven days in Taiwan - by Noah Smith - Noahpinion4 savers
- Lexica3 savers
- hesephist/monocle: Universal personal search engine, powered by a full text search algorithm written in pure Ink, indexing Linus's blogs and private note archives, contacts, tweets, and over a decade of journals.3 savers
- TOD-Bert2 savers
- I Will Teach You To Be Rich2 savers
- Distill Update 20182 savers
- 為何資料工程人才難以培養? – leafwind.tw2 savers
- THA LTD.2 savers
- PromptLayer - The first platform built for prompt engineers2 savers
- Tube Noise Map: The Noisiest Spots On The Underground Revealed2 savers
- microsoft/guidance: A guidance language for controlling large language models.2 savers
highlights — 401
In other words, I had been building many V1s, but none of them would have made the product stand out from another “GPT-wrapper”-SAAS.
Why are AI Products Doomed to Fail? | by Jeremy Arancio | Nov, 2023 | Towards AIBut I value it. I value it because I’m self-protective, because I’m insular, because I feel uneasy unless I know I’m doing in the right thing.
containmentDevelopers can now describe functions to gpt-4-0613 and gpt-3.5-turbo-0613, and have the model intelligently choose to output a JSON object containing arguments to call those functions.
Function calling and other API updatesSome people, however, insist on making grandiose demands on themselves, on others, and on life conditions and yet do not seem to be notably disturbed. Usually, they resort to self-deception, rationalizing, pollyannaism, withdrawal, distraction, supernaturalism, alcohol and drugs, relying on supporters, and desirelessness.
Must musturbation and demandingness lead to emotional disorders?This is an implementation detail, but it’s one that you can’t avoid for two reasons: APIs have token limits. If you try and send more than the limit you’ll get an error message like this one: “This model’s maximum context length is 4097 tokens. However, your messages resulted in 116142 tokens. Please reduce the length of the messages.” Tokens are how pricing works. gpt-3.5-turbo (the model used by ChatGPT, and the default model used by the llm command) costs $0.002 / 1,000 tokens. GPT-4 is $0.03 / 1,000 tokens of input and $0.06 / 1,000 for output. Being able to keep track of token counts is r…
llm, ttok and strip-tags—CLI tools for working with ChatGPT and other LLMsFor compiling our code, we are going to use Judge0. Judge0 is a simple, open-source code execution system that we can interact with.
How to Build a Code Editor with React that Compiles and Executes in 40+ Languageswe’ve used our custom d3 plugin called d3-horizon-charts to visualize data from all sensors for an entire month.
Location + time: urban data visualization - MORPHOCODEthree main types of temporal primitives: a time point; a time interval or a time span.
Location + time: urban data visualization - MORPHOCODEDramatron is a system that uses large language models that could be useful for authors for co-writing theatre scripts and screenplays. Dramatron uses hierarchical story generation for consistency across the generated text.
Dramatronwhen human programmers use functions and libraries for the first time, they frequently refer to textual resources such as code manuals and documentation, to explore and understand the available functionality. Inspired by this observation, we introduce DocPrompting: a natural-language-to-code generation approach that explicitly leverages documentation by (1) retrieving the relevant documentation pieces given an NL intent, and (2) generating code based on the NL intent and the retrieved documentation.
[2207.05987] DocPrompting: Generating Code by Retrieving the DocsAt deployment, ML or ops engineers need to translate these batch features into streaming features for the prediction pipeline and optimize them for latency.
Introduction to streaming for data scientistsThe actor aims to maximize the expected return R t . = ∑ ∞ τ =0 γ τ r t + τ with a discount factor γ = 0 . 997 for each model state. To consider rewards beyond the prediction horizon T = 16 , the critic learns to predict the return of each state under the current actor behavior: Actor: a t ∼ π θ ( a t | s t ) Critic: v ψ ( s t ) ≈ E p φ ,π θ [ R t ] (6) Starting from representations of replayed inputs, the dynamics predictor and actor produce a sequence of imagined model states s 1: T , actions a 1: T , rewards r 1: T , and continuation flags c 1: T .
[2301.04104] Mastering Diverse Domains through World ModelsThe actor and critic neural networks learn behaviors purely from abstract sequences predicted by the world mode
[2301.04104] Mastering Diverse Domains through World ModelsThe world model learns compact representations of sensory inputs through autoencoding 23,24 and enables planning by predicting future representations and rewards for potential actions
[2301.04104] Mastering Diverse Domains through World ModelsWe suggest symlog predictions as a simple solution to this dilemma.
[2301.04104] Mastering Diverse Domains through World ModelsPredicting large targets using a squared loss can lead to divergence whereas absolute and Huber losses 18 stagnate learning
[2301.04104] Mastering Diverse Domains through World ModelsThe DreamerV3 algorithm consists of 3 neural networks—the world model, the critic, and the actor—that are trained concurrently from replayed experience without sharing gradients,
[2301.04104] Mastering Diverse Domains through World Modelsowever, applying algorithms to new application domains, for example from board games to video games or robotics tasks, requires expert knowledge and computational resources for tuning the algorithms 3 . This brittleness also hinders scaling to large models that are expensive to tune
[2301.04104] Mastering Diverse Domains through World ModelsMy sense is that this individualism goes hand-in-hand with Taiwan’s general laid-back-ness, and its tolerance — when nobody is snapping at you to fall in line, you learn to pretty much do what you like.
Eleven days in Taiwan - by Noah Smith - NoahpinionTaiwan feels like a highly individualistic place
Eleven days in Taiwan - by Noah Smith - Noahpinionaccretion of remittance building projects over time shapes the social spaces of migration here (one point on the migration trajectory) and there (the second point on such trajectory),
AH110 - Material + Spatial Glocal Perspectivesaided by the nations’ geographic proximity and intertwined economies, institutionalized a back-and-forth or “circular migration.”
AH110 - Material + Spatial Glocal PerspectivesThus it remains unknown how immigration policy and immigration flows specifically influenced the building of, and people’s experiences of, remittance landscapes in Italy, Poland, or China at that time
AH110 - Material + Spatial Glocal Perspectivesby moving back and forth between two or more places, migrants build alliances “there” through “here,” essentially expanding the scope of “home” to include both “there” and “here.”
AH110 - Material + Spatial Glocal Perspectives. An architectural analysis of the spaces of their houses demonstrates remittance houses are a unit of analysis for larger social, political, and architectural discourses about migration and global building practices in rural localities.
AH110 - Material + Spatial Glocal Perspectivesfocuses on how migration is interwoven with the hopes and dreams of building a house in one’s hometown.
AH110 - Material + Spatial Glocal Perspectivesthe remittance house has crystallized migrant narratives and desires amid shifting cultural milieux. Artifacts of complex relationships, these houses are also embedded in the macro processes of globalization and transnational migration.
AH110 - Material + Spatial Glocal Perspectivesemphasize remitting and migration as key components of contemporary transnational building practices across the globe.
AH110 - Material + Spatial Glocal PerspectivesNestled in a poor Mexican town reliant on sugar cane and corn farming, this mansion is an exceptional example of what I call the “remittance house.” This term refers to a house built with money earned by a Mexican migrant in the United States who sends dollars—remits—to Mexico for the construction of his or her dream house
AH110 - Material + Spatial Glocal PerspectivesTo a large extent, the major business centers in the world today draw their importance from these transnational networks. There is no such entity as a single global city—and, in this sense, there is a sharp contrast with the erstwhile capitals of empires.
AH110 - Global CitiesGlobalization has given rise to new information technologies, the intensifying of transnational and translocal dynamics, and the strengthening presence and voice of sociocultural diversity.
AH110 - Global CitiesThese sociologists confronted massive processes—industrialization, urbanization, alienation, and a new cultural formation they called “urbanity.” Studying the city was not simply studying the urban. It was about studying the major social processes of an era.
AH110 - Global CitiesThe city and the metropolitan region have become locations where major macrosocial trends materialize and hence can be constituted as an object of global studies
AH110 - Global CitiesJacobs argued that utopian planners neglected to see the on-the-ground social relations that made city neighborhoods lively, vibrant places.
AH110 - Planning The CityRadiant City would facilitate and rationalize plans for mobility in the city, including multiple levels of roadway for automobile traffic.
AH110 - Planning The Cityr planned to decongest the center of the city by creating organized density at the heart of his cities.
AH110 - Planning The CityCity planning, according to LeCorbusier, was too important to be left in the hands of everyday people. Instead, he argued for the expertise of the city planner and the science of a profession to design cities.
AH110 - Planning The Citythe Chicago School drew on paradigms from the natural sciences to understand social processes in the city. Through a model of human ecology, they introduced concepts like invasion and succession, symbiosis, and dominance and growth.
AH110 - Planning The Citythe Chicago School used the city as a laboratory to understand the social processes of the modern metropolis.
AH110 - Planning The Citya group of sociologists at the University of Chicago began to explore the city as their urban laboratory. These sociologists, known as the Chicago School, emerged as the leading theorists of contemporary city life (~1915).
AH110 - Planning The CitySimmel suggests that the only way to survive everyday life in the contemporary city is through the adoption of a blasé attitude
AH110 - Planning The CityHoward identified nature, or “country” as the inseparable basis of physical and aesthetic expression in human society
AH110 - Planning The CityHoward proposed the garden city idea: unifying “town and country” in new and revitalized communities, as a central path toward resolving the mounting antagonisms of rapid urban growth; a garden city was to be set in nature as much as infused by nature throughout
AH110 - Planning The CityTheir visions of each was expressed, respectively, as the Garden City and the Radiant City.
AH110 - Planning The CityThe conditions of the industrializing city–the squalor described by Engels, for example–prompted the desire to plan cities.
AH110 - Planning The CityThe Industrial Revolution that called the people of Europe to its cities did not automatically foster enlightened and effective government planning and housing policies
AH110 - Planning The CityThe urbanite had become a hybrid, wholly dependent upon external conditions of supply and demand over which the individual had no control.
AH110 - Planning The CityWhat was new was the monotony of industrial work and the strict and constant supervision under which it was carried out. It may have been because the repetitive work was so dulling and dehumanizing—in addition to the fact that through the practice of child labor, adults had become socialized to accept their condition—that workers had neither the energy nor the will to successfully resist the imposition
AH110 - Planning The CityIt was the dimensions of this growth that led Davis (1955, 433) to conclude that “the transformation thus achieved in the nineteenth century was the true urban revolution, for it meant not only the rise of a few scattered towns and cities, but the appearance of genuine urbanization in the sense that a substantial proportion of the population lived in towns and cities.”
AH110 - The Ancient Citya number of other changes, having no direct relationship to industrial or manufacturing technology, were occurring at the same time, each seemingly amplifying the effect of the others.
AH110 - The Ancient City