Welcome to the Abacus.AI blog

Smaug-Decisions is an open 4B model that answers typed questions about audio, images and text with a calibrated probability per option, in one forward pass. It beats Gemini 4 Argon on audio decisions, and its 3.45 GB phone build runs offline on an iPhone GPU.

Today we are releasing Smaug Humanizer, a 12B model that rewrites AI-generated text so it reads like a person wrote it. It keeps the facts, the numbers and the layout of the input, and it...

We are releasing Smaug AI Image Detector, an open-source model that tells you whether an image was made by an AI image generator — the strongest open-source detector on real-world images, small enough to run on a laptop.

The Abacus.AI team has published five papers to appear at the Conference on Neural Information Processing Systems (NeurIPS) 2023. Three papers are in the main conference (including one awarded with an Oral Presentation) and two...

Today we release a larger 70B version of Giraffe, succeeding the 13B model we mentioned in a previous blog post. Giraffe is a family of models that are finetuned from base Llama 2 and use...

The Abacus.AI team has published two papers to appear at the Conference on Neural Information Processing Systems (NeurIPS) 2022. NeurIPS is a top machine learning conference held every December. Abacus.AI is also planning a social...

Today, we at Abacus.AI are thrilled to announce our Series A funding and Mike Volpi and Ram Shriram as Board Directors. In addition, we are open-sourcing our debiasing module. In this post, we give an introduction to bias in computer vision models and discuss our new research on debiasing models.

Neural architecture search (NAS) is a popular area of machine learning, with the goal of automating the development of the best neural network for a given dataset. In this post, we summarize our recent paper, which suggests that existing NAS benchmarks may be too small to effectively evaluate NAS algorithms.

In this post, we give an introduction to bias in machine learning, and we discuss new techniques for debiasing pretrained neural networks (i.e., post-hoc debiasing).