Efficient training of language models to fill in the middle
We show that an autoregressive language model can learn text embeddings after applying a simple transformation to the dataset. This simply moves a range of text from the middle of…
Scaling law for overoptimization of reward models
Reinforcement learning from human feedback typically optimizes against a reward model that has been trained to predict human preferences. Since the reward model is an imperfect proxy, overoptimizing its value…
How to predict and reduce the risk of potential abuse of language models in disinformation campaigns
As generative language models improve, they open up new possibilities in fields as diverse as medicine, law, education, and science. But like any new technology, it’s worth considering that they…
Early consideration of the potential impact of large-scale language models on the labor market
We investigate the potential impact of the Generative Pre-Trained Transformer (GPT) model and related technologies on the U.S. labor market. Use a new rubric to evaluate occupations based on their…
Language models can explain neurons using language models.
Although most of our explanations have low scores, we believe that we can further improve our ability to create explanations using ML techniques. For example, I found that you can…
General availability of GPT-4 API and deprecation of older models in Completions API
Applications that use the stable model name of the base GPT-3 model (ada, babbage, curie, davinci) will be automatically upgraded to the new model listed above on January 4, 2024.…
This AI paper reveals a new approach to understanding deep learning models: Using Concept Relevance Propagation (CRP) to figure out the ‘where’ and ‘what’
https://www.nature.com/articles/s42256-023-00711-8 The field of machine learning and artificial intelligence has become very important. We are making new advances every day. This area affects all spheres. Utilizing carefully developed neural network…
OpenAI partners with Scale to provide support for enterprise fine-tuning models
OpenAI and Scale are working together to enable more companies to benefit from fine-tuning our cutting-edge models. Enterprises expect high performance, ease of use, and customization when deploying AI into…
MIT’s machine learning system based on light could produce more powerful large-scale language models
by Elizabeth A. Thomson, MIT Materials Laboratory September 24, 2023 An artist’s rendition of a light-based computer system that could potentially power the power of machine learning programs like ChatGPT.…