# Chip Huyen I work to bring AI into production. I write about AI system design. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Common pitfalls when building generative AI applications DevFeed: [Common pitfalls when building generative AI applications]() Original publisher: [Read original article]() Author: Chip Huyen Published: 2025-01-16T00:00:00Z Content type: opinion Language: en Sources: [Chip Huyen]() Topics: [Generative AI](), [foundation-models](), [AI Development](), [Large Language Model](), [Optimization]() Tags: [foundation-models](), [generative-ai](), [llm](), [mistakes](), [optimization](), [pitfalls]() ### AI overview This article discusses common pitfalls in building applications with foundation models. It cautions against using generative AI where simpler optimization methods may be cheaper and more reliable, and distinguishes solving a problem from merely testing whether generative AI can be applied. ### Source excerpt As we're still in the early days of building applications with foundation models, it's normal to make mistakes. This is a quick note with examples of some of the most common pitfalls that I've seen, both from public case studies and from my personal experience. Because these pitfalls are common, if you've worked on any AI product, you've probably seen them before. 1. Use generative AI when you don't need generative AI Every time there's a new technology, I can hear the collective sigh of senior engineers everywhere: "Not everything is a nail." Generative AI isn't an exception -- its seemingly limitless capabilities only exacerbate the tendency to use generative AI for everything. A team pitched me the idea of using generative AI to optimize energy consumption. They fed a household's list of energy-intensive activities and hourly electricity prices into an LLM, then asked it to create a schedule to minimize energy costs. Their experiments showed that this could help reduce a household's electricity bill by 30%. Free money. Why wouldn't anyone want to use their app? I asked: "How does it compare to simply scheduling the most energy-intensive activities when electricity is cheapest? Say, doing your laundry and charging your car after 10pm?" They said they would try it later and let me know. They never followed up, but they abandoned this app soon after. I suspect that this greedy scheduling can be quite effective. Even if it's not, there are other much cheaper and more reliable optimization solutions than generative AI, like linear programming. I've seen this scenario over and over again. A big company wants to use generative AI to detect anomalies in network traffic. Another wants to predict upcoming customer call volume. A hospital wants to detect whether a patient is malnourished (really not recommended). It can often be beneficial to explore a new approach to get a sense of what's possible, as long as you're aware that your goal isn't to solve a problem but to test ## Agents DevFeed: [Agents]() Original publisher: [Read original article]() Author: Chip Huyen Published: 2025-01-07T00:00:00Z Content type: article Language: en Sources: [Chip Huyen]() Topics: [Artificial Intelligence](), [Tool](), [foundation-models](), [AI Engineering](), [AI Development]() Tags: [agents](), [ai](), [ai-engineering](), [foundation-models](), [tools]() ### AI overview This article provides an overview of intelligent agents, focusing on how foundation models enable agentic applications. It examines tools and planning as factors that shape agent capabilities, and discusses agent failure modes and evaluation. The author notes that the field is emerging and lacks established theoretical frameworks. ### Source excerpt Intelligent agents are considered by many to be the ultimate goal of AI. The classic book by Stuart Russell and Peter Norvig, Artificial Intelligence: A Modern Approach (Prentice Hall, 1995), defines the field of AI research as "the study and design of rational agents." The unprecedented capabilities of foundation models have opened the door to agentic applications that were previously unimaginable. These new capabilities make it finally possible to develop autonomous, intelligent agents to act as our assistants, coworkers, and coaches. They can help us create a website, gather data, plan a trip, do market research, manage a customer account, automate data entry, prepare us for interviews, interview our candidates, negotiate a deal, etc. The possibilities seem endless, and the potential economic value of these agents is enormous. This section will start with an overview of agents and then continue with two aspects that determine the capabilities of an agent: tools and planning. Agents, with their new modes of operations, have new modes of failure. This section will end with a discussion on how to evaluate agents to catch these failures. This post is adapted from the Agents section of AI Engineering (2025) with minor edits to make it a standalone post. Notes: AI-powered agents are an emerging field with no established theoretical frameworks for defining, developing, and evaluating them. This section is a best-effort attempt to build a framework from the existing literature, but it will evolve as the field does. Compared to the rest of the book, this section is more experimental. I received helpful feedback from early reviewers, and I hope to get feedback from readers of this blog post, too. Just before this book came out, Anthropic published a blog post on Building effective agents (Dec 2024). I'm glad to see that Anthropic's blog post and my agent section are conceptually aligned, though with slightly different terminologies. However, Anthropic's post focuses on iso ## Building A Generative AI Platform DevFeed: [Building A Generative AI Platform]() Original publisher: [Read original article]() Author: Chip Huyen Published: 2024-07-25T00:00:00Z Content type: tutorial Language: en Sources: [Chip Huyen]() Topics: [AI Platform](), [Generative AI](), [Architecture & Design](), [AI Development](), [API](), [Latency](), [Caching](), [Orchestration](), [observability](), [Security]() Tags: [ai](), [ai-platform](), [apis](), [architecture](), [cache](), [gateway](), [generative-ai](), [latency](), [observability](), [orchestration](), [security]() ### AI overview This post explains the common architecture of a generative AI platform, starting with a simple application that sends queries to a model and progressively adding components such as context construction, guardrails, model routers and gateways, caching, complex logic, observability, and orchestration. It notes that the architecture should be adapted to each system and that evaluation is needed throughout development. ### Source excerpt After studying how companies deploy generative AI applications, I noticed many similarities in their platforms. This post outlines the common components of a generative AI platform, what they do, and how they are implemented. I try my best to keep the architecture general, but certain applications might deviate. This is what the overall architecture looks like. This is a pretty complex system. This post will start from the simplest architecture and progressively add more components. In its simplest form, your application receives a query and sends it to the model. The model generates a response, which is returned to the user. There are no guardrails, no augmented context, and no optimization. The Model API box refers to both third-party APIs (e.g., OpenAI, Google, Anthropic) and self-hosted APIs. From this, you can add more components as needs arise. The order discussed in this post is common, though you don't need to follow the exact same order. A component can be skipped if your system works well without it. Evaluation is necessary at every step of the development process. Enhance context input into a model by giving the model access to external data sources and tools for information gathering. Put in guardrails to protect your system and your users. Add model router and gateway to support complex pipelines and add more security. Optimize for latency and costs with cache. Add complex logic and write actions to maximize your system's capabilities. Observability, which allows you to gain visibility into your system for monitoring and debugging, and orchestration, which involves chaining all the components together, are two essential components of the platform. We will discuss them at the end of this post. " What this post is not " This post focuses on the overall architecture for deploying AI applications. It discusses what components are needed and considerations when building these components. It's not about how to build AI applications and, therefore, does NOT di ## What I learned from looking at 900 most popular open source AI tools DevFeed: [What I learned from looking at 900 most popular open source AI tools]() Original publisher: [Read original article]() Author: Chip Huyen Published: 2024-03-14T00:00:00Z Content type: article Language: en Sources: [Chip Huyen]() Topics: [Artificial Intelligence](), [Open Source](), [foundation-models](), [Development](), [Large Language Model](), [AI Engineering](), [Machine learning](), [Application Development](), [GitHub](), [Prompt Engineering](), [Retrieval Augmented Generation (RAG)](), [PyTorch]() Tags: [ai](), [engineering](), [foundation-models](), [github](), [model](), [open-source](), [open-source-ai](), [pytorch](), [skypilot](), [transformers](), [vector]() ### AI overview The article analyzes 896 popular open-source AI repositories, focusing on 845 software repositories after excluding tutorials and aggregated lists. It organizes the ecosystem around foundation models into infrastructure, model development, and application development, and notes differences between China's open-source ecosystem and Western ecosystems. ### Source excerpt [Hacker News discussion, LinkedIn discussion, Twitter thread] Update (Feb 2026): The full list of open source AI repos is hosted at Good AI List, updated daily. It's balooned to 15K repos, and you can submit missing repos. You can also find some of them on my cool-llm-repos list on GitHub. Four years ago, I did an analysis of the open source ML ecosystem. Since then, the landscape has changed, so I revisited the topic. This time, I focused exclusively on the stack around foundation models. Data I searched GitHub using the keywords gpt, llm, and generative ai. If AI feels so overwhelming right now, it's because it is. There are 118K results for gpt alone. To make my life easier, I limited my search to the repos with at least 500 stars. There were 590 results for llm, 531 for gpt, and 38 for generative ai. I also occasionally checked GitHub trending and social media for new repos. After MANY hours, I found 896 repos. Of these, 51 are tutorials (e.g. dair-ai/Prompt-Engineering-Guide) and aggregated lists (e.g. f/awesome-chatgpt-prompts). While these tutorials and lists are helpful, I'm more interested in software. I still include them in the final list, but the analysis is done with the 845 software repositories. It was a painful but rewarding process. It gave me a much better understanding of what people are working on, how incredibly collaborative the open source community is, and just how much China's open source ecosystem diverges from the Western one. The New AI Stack I think of the AI stack as consisting of 3 layers: infrastructure, model development, and application development. Infrastructure At the bottom is the stack is infrastructure, which includes toolings for serving (vllm, NVIDIA's Triton), compute management (skypilot), vector search and database (faiss, milvus, qdrant, lancedb), .... Model development This layer provides toolings for developing models, including frameworks for modeling & training (transformers, pytorch, DeepSpeed), inference optimization ## Predictive Human Preference: From Model Ranking to Model Routing DevFeed: [Predictive Human Preference: From Model Ranking to Model Routing]() Original publisher: [Read original article]() Author: Chip Huyen Published: 2024-02-28T00:00:00Z Content type: article Language: en Sources: [Chip Huyen]() Topics: [Model Routing](), [AI Models](), [Model Development](), [post-training](), [AI Chat]() Tags: [ai-models](), [anthropic](), [chatgpt](), [dpo](), [gemini](), [latency](), [model](), [model-routing](), [ranking](), [rlhf]() ### AI overview The article examines predictive human preference: predicting which AI model users will prefer for a specific prompt. It describes model routing as a use case, where prompts could be directed to a preferred model to potentially reduce cost and latency, and discusses using preference predictions to interpret model strengths and weaknesses. It also outlines evaluating predictions against Chatbot Arena and building a preference predictor. ### Source excerpt A challenge of building AI applications is choosing which model to use. What if we don't have to? What if we can predict the best model for any prompt? Predictive human preference aims to predict which model users might prefer for a specific query. Human preference has emerged to be both the Northstar and a powerful tool for AI model development. Human preference guides post-training techniques including RLHF and DPO. Human preference is also used to rank AI models, as used by LMSYS's Chatbot Arena. Chatbot Arena aims to determine which model is generally preferred. I wanted to see if it's possible to predict which model is preferred for each query. One use case of predictive human preference is model routing. For example, if we know in advance that for a prompt, users will prefer Claude Instant's response over GPT-4, and Claude Instant is cheaper/faster than GPT-4, we can route this prompt to Claude Instant. Model routing has the potential to increase response quality while reducing costs and latency. Another use case of predictive human preference is interpretability. Mapping out a model's performance on different prompts can help us understand this model's strengths and weaknesses. See section Experiment results for examples. Here's what predictive human preference for different model pairs looks like for the prompt "What's the best way to cluster text embeddings?". The predictions were generated by my toy preference predictor. The bright yellow color for the (GPT-4, GPT-3.5-Turbo) cell means that my predictor thinks GPT-4's response is very likely to be preferred to that of GPT-3.5-Turbo's for this prompt. This post first discusses the correctness of Chatbot Arena, which will then be used as a baseline to evaluate the correctness of preference predictions. It then discusses how to build a preference predictor and the initial results. Ranking Models Using Human Preference Using preferential signals (comparisons) to rank models has grown in popularity in the last ## Generation configurations: temperature, top-k, top-p, and test time compute DevFeed: [Generation configurations: temperature, top-k, top-p, and test time compute]() Original publisher: [Read original article]() Author: Chip Huyen Published: 2024-01-16T00:00:00Z Content type: tutorial Language: en Sources: [Chip Huyen]() Topics: [Artificial Intelligence](), [Machine learning](), [Language models](), [Neural Network](), [Inference]() Tags: [ai](), [decoding](), [hallucinations](), [inference](), [ml](), [neural](), [neural-network](), [responses](), [token](), [tokens]() ### AI overview This tutorial explains why machine-learning models produce probabilistic responses and how sampling, or decoding, generates them. It covers sampling strategies such as temperature, top-k, and top-p, test-time compute through multiple outputs, and structured outputs. ### Source excerpt ML models are probabilistic. Imagine that you want to know what's the best cuisine in the world. If you ask someone this question twice, a minute apart, their answers both times should be the same. If you ask a model the same question twice, its answer can change. If the model thinks that Vietnamese cuisine has a 70% chance of being the best cuisine and Italian cuisine has a 30% chance, it'll answer "Vietnamese" 70% of the time, and "Italian" 30%. This probabilistic nature makes AI great for creative tasks. What is creativity but the ability to explore beyond the common possibilities, to think outside the box? However, this probabilistic nature also causes inconsistency and hallucinations. It's fatal for tasks that depend on factuality. Recently, I went over 3 months' worth of customer support requests of an AI startup I advise and found that ⅕ of the questions are because users don't understand or don't know how to work with this probabilistic nature. To understand why AI's responses are probabilistic, we need to understand how models generate responses, a process known as sampling (or decoding). This post consists of 3 parts. Sampling: sampling strategies and sampling variables including temperature, top-k, and top-p. Test time compute: increasing the compute allocated to inference, e.g. sampling multiple outputs, to help improve a model's performance. Structured outputs: how to get models to generate outputs in a certain format. Sampling Given an input, a neural network produces an output by first computing the probabilities of all possible values. For a classifier, possible values are the available classes. For example, if a model is trained to classify whether an email is spam, there are only two possible values: spam and not spam. The model computes the probability of each of these two values, say being spam is 90% and not spam is 10%. To generate the next token, a language model first computes the probability distribution over all tokens in the vocabulary. Fo ## Multimodality and Large Multimodal Models (LMMs) DevFeed: [Multimodality and Large Multimodal Models (LMMs)]() Original publisher: [Read original article]() Author: Chip Huyen Published: 2023-10-10T00:00:00Z Content type: tutorial Language: en Sources: [Chip Huyen]() Topics: [multimodal](), [AI Research](), [Large Language Model]() Tags: [ai](), [ai-research](), [models](), [multimodal]() ### AI overview This tutorial explains multimodal systems and Large Multimodal Models (LMMs), including their data modalities, task types, foundational approaches such as CLIP and Flamingo, and research areas such as multimodal outputs and adapter-based training. ### Source excerpt For a long time, each ML model operated in one data mode - text (translation, language modeling), image (object detection, image classification), or audio (speech recognition). However, natural intelligence is not limited to just a single modality. Humans can read, talk, and see. We listen to music to relax and watch out for strange noises to detect danger. Being able to work with multimodal data is essential for us or any AI to operate in the real world. OpenAI noted in their GPT-4V system card that "incorporating additional modalities (such as image inputs) into LLMs is viewed by some as a key frontier in AI research and development." Incorporating additional modalities to LLMs (Large Language Models) creates LMMs (Large Multimodal Models). Not all multimodal systems are LMMs. For example, text-to-image models like Midjourney, Stable Diffusion, and Dall-E are multimodal but don't have a language model component. Multimodal can mean one or more of the following: Input and output are of different modalities (e.g. text-to-image, image-to-text) Inputs are multimodal (e.g. a system that can process both text and images) Outputs are multimodal (e.g. a system that can generate both text and images) This post covers multimodal systems in general, including LMMs. It consists of 3 parts. Part 1 covers the context for multimodality, including why multimodal, different data modalities, and types of multimodal tasks. Part 2 discusses the fundamentals of a multimodal system, using the examples of CLIP, which lays the foundation for many future multimodal systems, and Flamingo, whose impressive performance gave rise to LMMs. Part 3 discusses some active research areas for LMMs, including generating multimodal outputs and adapters for more efficient multimodal training, covering newer multimodal systems such as BLIP-2, LLaVA, LLaMA-Adapter V2, LAVIN, etc. The post is long. Feel free to skip to the sections most interesting to you. ⚠ Ambiguous terminology ⚠ Multimodal data can als ## Open challenges in LLM research DevFeed: [Open challenges in LLM research]() Original publisher: [Read original article]() Author: Chip Huyen Published: 2023-08-16T00:00:00Z Content type: opinion Language: en Sources: [Chip Huyen]() Topics: [Large Language Model](), [Hallucination detection](), [context](), [LLM evaluation / benchmarking](), [Chain-of-thought](), [multimodal]() Tags: [ai](), [chain-of-thought](), [challenges](), [context](), [evaluation](), [hallucination-detection](), [llm](), [multimodal](), [research]() ### AI overview This article surveys ten research directions aimed at improving large language models. The supplied excerpt focuses on reducing and measuring hallucinations, optimizing context length and construction, and briefly identifies multimodality, new architectures, and GPU alternatives as areas of interest. ### Source excerpt [LinkedIn discussion, Twitter thread] Never before in my life had I seen so many smart people working on the same goal: making LLMs better. After talking to many people working in both industry and academia, I noticed the 10 major research directions that emerged. The first two directions, hallucinations and context learning, are probably the most talked about today. I'm the most excited about numbers 3 (multimodality), 5 (new architecture), and 6 (GPU alternatives). 1. Reduce and measure hallucinations Hallucination is a heavily discussed topic already so I'll be quick. Hallucination happens when an AI model makes stuff up. For many creative use cases, hallucination is a feature. However, for most other use cases, hallucination is a bug. I was at a panel on LLM with Dropbox, Langchain, Elastics, and Anthropic recently, and the #1 roadblock they see for companies to adopt LLMs in production is hallucination. Mitigating hallucination and developing metrics to measure hallucination is a blossoming research topic, and I've seen many startups focus on this problem. There are also ad-hoc tips to reduce hallucination, such as adding more context to the prompt, chain-of-thought, self-consistency, or asking your model to be concise in its response. To learn more about hallucination: Survey of Hallucination in Natural Language Generation (Ji et al., 2022) How Language Model Hallucinations Can Snowball (Zhang et al., 2023) A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity (Bang et al., 2023) Contrastive Learning Reduces Hallucination in Conversations (Sun et al., 2022) Self-Consistency Improves Chain of Thought Reasoning in Language Models (Wang et al., 2022) SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models (Manakul et al., 2023) A simple example of fact-checking and hallucination by NVIDIA's NeMo-Guardrails 2. Optimize context length and context construction A vast ma ## Generative AI Strategy DevFeed: [Generative AI Strategy]() Original publisher: [Read original article]() Author: Chip Huyen Published: 2023-06-07T00:00:00Z Content type: opinion Language: en Sources: [Chip Huyen]() Topics: [AI Strategy](), [Generative AI]() Tags: [ai](), [ai-strategy](), [generative-ai](), [strategy]() ### AI overview A short post introduces a talk for leaders trying to determine what to do with generative AI. It presents the talk as a simple framework for exploring generative AI strategy, while noting that the ideas are still being developed. ### Source excerpt I had a lot of fun preparing the talk: "Leadership needs us to do generative AI. What do we do?" for Fully Connected. The idea for the talk came from many conversations I've had recently with friends who need to figure out their generative AI strategy, but aren't sure what exactly to do. This talk is a simple framework to explore what to do with generative AI. Many ideas are still being fleshed out. I hope to convert this into a proper post when I have more time. In the meantime, I'd love to hear from your experience through this process. I couldn't figure out how to make the slides centered on the page. You might want to download the slides. Thanks everyone who responded to my post and shared your thoughts on what I should include in the talk. Thanks Kyle Gallatin, Goku Mohandas, Han-chung Lee, and Jamie de Guerre for thoughtful feedback on the talk.