LWHarness Engineering for Self-ImprovementThe concept of recursive self-improvement (RSI) dates back to I. J. Good (1965) , where he defined an “ultraintelligent machine” as a system that can surpass humans in all intellectual activities and design better machines to imLWLil'Log2026/07/04 08:00
LWScaling Laws, CarefullyScaling laws are one of the most critical empirical findings in deep learning. The observation is simple in form: the training loss $L$ decreases predictably as we scale up model size $N$, dataset size $D$, and compute $C$, following a poweLWLil'Log2026/06/24 08:00
LWWhy We ThinkSpecial thanks to John Schulman for a lot of super valuable feedback and direct edits on this post. Test time compute ( Graves et al. 2016 , Ling, et al. 2017 , Cobbe et al. 2021 ) and Chain-of-thought (CoT) ( Wei et al. 2022 , Nye et al. 2LWLil'Log2025/05/01 08:00
LWReward Hacking in Reinforcement LearningReward hacking occurs when a reinforcement learning (RL) agent exploits flaws or ambiguities in the reward function to achieve high rewards, without genuinely learning or completing the intended task. Reward hacking exists because RL enviroLWLil'Log2024/11/28 08:00
LWExtrinsic Hallucinations in LLMsHallucination in large language models usually refers to the model generating unfaithful, fabricated, inconsistent, or nonsensical content. As a term, hallucination has been somewhat generalized to cases when the model makes mistakes. Here,LWLil'Log2024/07/07 08:00
LWDiffusion Models for Video GenerationDiffusion models have demonstrated strong results on image synthesis in past years. Now the research community has started working on a harder task—using it for video generation. The task itself is a superset of the image case, since LWLil'Log2024/04/12 08:00
LWThinking about High-Quality Human Data[Special thank you to Ian Kivlichan for many useful pointers (E.g. the 100+ year old Nature paper “Vox populi”) and nice feedback. 🙏 ] High-quality data is the fuel for modern data deep learning model training. Most of the taskLWLil'Log2024/02/05 08:00
LWAdversarial Attacks on LLMsThe use of large language models in the real world has strongly accelerated by the launch of ChatGPT. We (including my team at OpenAI, shoutout to them) have invested a lot of effort to build default safe behavior into the model during the LWLil'Log2023/10/25 08:00
LLM Powered Autonomous AgentsBuilding agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT , GPT-Engineer and BabyAGI , serve as inspiring examples. The potentiality of LLM extends beyond geneLWLil'Log2023/06/23 08:00
LWPrompt EngineeringPrompt Engineering , also known as In-Context Prompting , refers to methods for how to communicate with LLM to steer its behavior for desired outcomes without updating the model weights. It is an empirical science and the effect of prompt eLWLil'Log2023/03/15 08:00
LWThe Transformer Family Version 2.0Many new Transformer architecture improvements have been proposed since my last post on “The Transformer Family” about three years ago. Here I did a big refactoring and enrichment of that 2020 post — restructure the hierarLWLil'Log2023/01/27 08:00
LWLarge Transformer Model Inference Optimization[Updated on 2023-01-24: add a small section on Distillation .] Large transformer models are mainstream nowadays, creating SoTA results for a variety of tasks. They are powerful but very expensive to train and use. The extremely high inferenLWLil'Log2023/01/11 01:00
LWSome Math behind Neural Tangent KernelNeural networks are well known to be over-parameterized and can often easily fit data with near-zero training loss with decent generalization performance on test dataset. Although all these parameters are initialized at random, the optimizaLWLil'Log2022/09/09 01:00
LWGeneralized Visual Language ModelsProcessing images to generate text, such as image captioning and visual question-answering, has been studied for years. Traditionally such systems rely on an object detection network as a vision encoder to capture visual features and then pLWLil'Log2022/06/10 06:10
LWLearning with not Enough Data Part 3: Data GenerationHere comes the Part 3 on learning with not enough data (Previous: Part 1 and Part 2 ). Let’s consider two approaches for generating synthetic data for training. Augmented data . Given a set of existing training samples, we can apply a varieLWLil'Log2022/04/16 06:10
LWLearning with not Enough Data Part 2: Active LearningThis is part 2 of what to do when facing a limited amount of labeled data for supervised learning tasks. This time we will get some amount of human labeling work involved, but within a budget limit, and therefore we need to be smart when seLWLil'Log2022/02/20 08:00
LWLearning with not Enough Data Part 1: Semi-Supervised LearningWhen facing a limited amount of labeled data for supervised learning tasks, four approaches are commonly discussed.LWLil'Log2021/12/05 08:00
LWHow to Train Really Large Models on Many GPUs?[Updated on 2022-03-13: add expert choice routing .] [Updated on 2022-06-10]: Greg and I wrote a shorted and upgraded version of this post, published on OpenAI Blog: “Techniques for Training Large Neural Networks”LWLil'Log2021/09/24 08:00
LWWhat are Diffusion Models?[Updated on 2021-09-19: Highly recommend this blog post on score-based generative modeling by Yang Song (author of several key papers in the references)]. [Updated on 2022-08-27: Added classifier-free guidance , GLIDE , unCLIP and Imagen . LWLil'Log2021/07/11 08:00
LWContrastive Representation LearningThe goal of contrastive representation learning is to learn such an embedding space in which similar sample pairs stay close to each other while dissimilar ones are far apart. Contrastive learning can be applied to both supervised and unsupLWLil'Log2021/05/31 08:00
今日热点