Annotated talk slides by Arvind Narayanan and Sayash Kapoor of Princeton University, dated October 4, 2023, on evaluating large language models. The slides discuss three hard problems in LLM evaluation: prompt sensitivity, construct validity and contamination, and examine questions about ChatGPT's behavior over time and its political bias.
Optuna is an open source automatic hyperparameter optimization software framework designed for machine learning. It provides an imperative, define-by-run style user API with eager search spaces, algorithms that prune unpromising trials, and parallelization of searches over multiple threads or processes without modifying code. It is framework agnostic and can be installed with pip on Python 3.10 or newer.
DiamantAI is a Substack newsletter authored by Nir Diamant that covers artificial intelligence. It publishes latest techniques, breakthroughs, insights, and unique tutorials for its over 40,000 subscribers. The newsletter is described as a top 0.1% AI resource and also maintains a GitHub repository on advanced RAG techniques with thousands of stars.
Qwen is an AI project whose site lists models including Qwen3Guard, a safety guardrail model built on Qwen3 foundation models for detecting risk in prompts and responses. It also lists Qwen-Image-Edit, an image editing version of a 20B Qwen-Image model, and Qwen-MT, a translation model via Qwen API supporting 92 languages.
TextBlob is a Python library for processing textual data. It offers a simple API for natural language processing tasks including part-of-speech tagging, noun phrase extraction, sentiment analysis, classification, and tokenization. The documentation lists features such as word and phrase frequencies, parsing n-grams, lemmatization, spelling correction, and WordNet integration. It builds on NLTK and pattern, and new models or languages can be added through extensions.
Distill publishes articles about machine learning, including peer-reviewed explanations of neural network topics such as graph neural networks, feature visualization and feature-wise transformations. In a July 2, 2021 editorial, the team announced that after five years Distill would be taking a break.
A Neural Network Playground is a browser-based educational visualization created by Daniel Smilkov and Shan Carter for experimenting with a real neural network. Users adjust controls to explore how a network learns from data. The project is open sourced on GitHub under the Apache License so its features can be tailored to a topic or lesson.
Community notes
Be the first to share how this project worked for you.
Sign in with a verified account to leave a review or join the discussion.
Sign in to reviewKeep exploring
Optuna is an open source automatic hyperparameter optimization software framework designed for machine learning. It provides an imperative, define-by-run style user API with eager search spaces, algorithms that prune unpromising trials, and parallelization of searches over multiple threads or processes without modifying code. It is framework agnostic and can be installed with pip on Python 3.10 or newer.
DiamantAI is a Substack newsletter authored by Nir Diamant that covers artificial intelligence. It publishes latest techniques, breakthroughs, insights, and unique tutorials for its over 40,000 subscribers. The newsletter is described as a top 0.1% AI resource and also maintains a GitHub repository on advanced RAG techniques with thousands of stars.
Qwen is an AI project whose site lists models including Qwen3Guard, a safety guardrail model built on Qwen3 foundation models for detecting risk in prompts and responses. It also lists Qwen-Image-Edit, an image editing version of a 20B Qwen-Image model, and Qwen-MT, a translation model via Qwen API supporting 92 languages.
TextBlob is a Python library for processing textual data. It offers a simple API for natural language processing tasks including part-of-speech tagging, noun phrase extraction, sentiment analysis, classification, and tokenization. The documentation lists features such as word and phrase frequencies, parsing n-grams, lemmatization, spelling correction, and WordNet integration. It builds on NLTK and pattern, and new models or languages can be added through extensions.
Distill publishes articles about machine learning, including peer-reviewed explanations of neural network topics such as graph neural networks, feature visualization and feature-wise transformations. In a July 2, 2021 editorial, the team announced that after five years Distill would be taking a break.
A Neural Network Playground is a browser-based educational visualization created by Daniel Smilkov and Shan Carter for experimenting with a real neural network. Users adjust controls to explore how a network learns from data. The project is open sourced on GitHub under the Apache License so its features can be tailored to a topic or lesson.
