We propose a set of precise criteria for saying a neural net learns and uses a "world model." The goal is to give an operational meaning to terms that are often used informally, in order to provide a common language for experimental investigation. We focus specifically on the idea of representing a latent "state space" of the world, leaving modeling the effect of actions to future work. Our definition is based on ideas from the linear probing literature, and formalizes the notion of a computation that factors through a representation of the data generation process. An essential addition to the definition is a set of conditions to check that such a "world model" is not a trivial consequence of the neural net's data or task.

Most of today's artificial intelligence (AI) systems rely on machine learning algorithms that can predict specific outcomes by drawing on pre-established values, but now researchers from OpenAI, a company funded by no less than Elon Musk and Peter Thiel, who are trying to democratise AI for "human good" just discovered – literally – that a machine learning system they created to predict the next character in the text of reviews from Amazon evolved into an unsupervised learning system that could learn how to read sentiment. That's a pretty big deal, and it's also something that, at the moment, even the researchers themselves can't explain. "We were very surprised that our model learned an interpretable feature, and that simply predicting the next character in Amazon reviews resulted in discovering the concept of sentiment," said OpenAI in a blog. According to the post OpenAI's neural network was able to train itself and analyse sentiment accurately by classifying Amazon's reviews as either positive or negative – and it then generated follow on text that fit with the sentiment. The AI the team used was what's known as a multiplicative long short-term memory (LSTM) model that was trained for a month, processing 12,500 characters a second using Nvidia Pascal GPU's – which Nvidia's own CEO gifted to Elon Musk last year – with "4,096 units on a corpus of 82 million Amazon reviews to predict the next character in a chunk of text."

Most of today's artificial intelligence (AI) systems rely on machine learning algorithms that can predict specific outcomes by drawing on pre-established values, but now researchers from OpenAI, a company funded by no less than Elon Musk and Peter Thiel, who are trying to democratise AI for "human good" just discovered – literally – that a machine learning system they created to predict the next character in the text of reviews from Amazon evolved into an unsupervised learning system that could learn how to read sentiment. That's a pretty big deal, and it's also something that, at the moment, even the researchers themselves can't explain. "We were very surprised that our model learned an interpretable feature, and that simply predicting the next character in Amazon reviews resulted in discovering the concept of sentiment," said OpenAI in a blog. According to the post OpenAI's neural network was able to train itself and analyse sentiment accurately by classifying Amazon's reviews as either positive or negative – and it then generated follow on text that fit with the sentiment. The AI the team used was what's known as a multiplicative long short-term memory (LSTM) model that was trained for a month, processing 12,500 characters a second using Nvidia Pascal GPU's – which Nvidia's own CEO gifted to Elon Musk last year – with "4,096 units on a corpus of 82 million Amazon reviews to predict the next character in a chunk of text."

Our system beats other approaches on Stanford Sentiment Treebank while using dramatically less data. The number of labeled examples it takes two variants of our model (the green and blue lines) to match fully supervised approaches, each trained with 6,920 examples (the dashed gray lines). Our L1-regularized model (pretrained in an unsupervised fashion on Amazon reviews) matches multichannel CNN performance with only 11 labeled examples, and state-of-the-art CT-LSTM Ensembles with 232 examples. We were very surprised that our model learned an interpretable feature, and that simply predicting the next character in Amazon reviews resulted in discovering the concept of sentiment. We believe the phenomenon is not specific to our model, but is instead a general property of certain large neural networks that are trained to predict the next step or dimension in their inputs.