Beyond Autoregression: A New Model For Text Generation

There are also trade-offs in creativity. Because the energy critic favors low-energy (i.e., high-probability) text, the model ...

How Intuit built custom financial LLMs that cut latency 50% while boosting accuracy—and what enterprise AI teams can learn

Intuit has been on a multi-year journey building its Generative AI Operating System (GenOS) to power AI agents across its platforms which include TurboTax, QuickBooks, Credit Karma and Mailchimp. The ...
Unite.AI

Research Finds Even a Little Bad Data Can Wreck a Fine-Tuned AI

A new study shows that fine-tuning ChatGPT on even small amounts of bad data can make it unsafe, unreliable, and veer it wildly off-topic. Just 10% of wrong answers in training data begins to break ...

DeepMind AI safety report explores the perils of “misaligned” AI

DeepMind's safety framework is based on so-called "critical capability levels" (CCLs). These are essentially risk assessment rubrics that aim to measure an AI model's capabilities and define the point ...
eLife

Importance of higher-order epistasis in protein sequence-function relationships

In this important work, the authors present a new transformer-based neural network designed to isolate and quantify higher-order epistasis in protein sequences. They provide solid evidence that higher ...