Demos

These are some of the open-source demos we built to make our research more readily available for developers and researchers.

	Predicting Multiplication without Intermediate Reasoning Steps Directly predict the product of 20-digit multiplication without chain-of-thought reasoning, using a 12-layer GPT-2. demo code
	Solving Grade School Math Problems without Intermediate Reasoning Steps Directly predict the answer to grade school math problems without chain-of-thought reasoning at 52% accuracy. demo code
	WildChat Visualizer (What Do People Use ChatGPT For?) A search tool for exploring 1M real user-ChatGPT conversations in WildChat. demo data
	AK Hugging Face Paper Selection Predictor Predict if @_akhaliq will select a paper into Hugging Face papers based on its title, authors, and abstract. demo code data models
	OpenAI Watch Monitoring the consistency of GPT-4 under greedy decoding (T=0) with a web-based application that commands GPT-4 to draw a unicorn every hour. demo data
	Markup-to-Image Diffusion Models with Scheduled Sampling A learning-based system to compile presentational markup, such as LaTeX, into corresponding images. demo code paper models
	Neural Linguistic Steganography Hide secret messages in fluent texts using arithmetic coding and strong neural models. demo code paper
	OpenNMT A full service open-source neural machine translation system. Originally developed in Lua with Systran, since ported to PyTorch and TensorFlow and maintained externally. code paper
	Image-to-Markup A learning-based system to decompile an image into presentational markup. For example, we can infer the LaTeX or HTML source from a rendered image. demo code paper data models
	Cascaded Text Generation with Markov Transformers A parallel, fast, autoregressive, and accurate text generation algorithm using high-order Conditional Random Fields (CRFs). code paper