![Mathematics | Free Full-Text | ADDA: An Adversarial Direction-Guided Decision-Based Attack via Multiple Surrogate Models Mathematics | Free Full-Text | ADDA: An Adversarial Direction-Guided Decision-Based Attack via Multiple Surrogate Models](https://www.mdpi.com/mathematics/mathematics-11-03613/article_deploy/html/images/mathematics-11-03613-g001.png)
Mathematics | Free Full-Text | ADDA: An Adversarial Direction-Guided Decision-Based Attack via Multiple Surrogate Models
![Cameron R. Wolfe, Ph.D. on X: "Next-token prediction is the workhorse behind all modern advancements in large language models (LLMs) due to its use in training these models over unlabeled text. But, Cameron R. Wolfe, Ph.D. on X: "Next-token prediction is the workhorse behind all modern advancements in large language models (LLMs) due to its use in training these models over unlabeled text. But,](https://pbs.twimg.com/media/FyxbmrTWIAARVYW.jpg:large)
Cameron R. Wolfe, Ph.D. on X: "Next-token prediction is the workhorse behind all modern advancements in large language models (LLMs) due to its use in training these models over unlabeled text. But,
What happens if predicted probabilities from different models don't agree with each other's on the same data set? Do the models still have good results or are they bad as well? -
![Comparison of model-predicted timelines from model-only simulations and... | Download Scientific Diagram Comparison of model-predicted timelines from model-only simulations and... | Download Scientific Diagram](https://www.researchgate.net/publication/328150678/figure/fig3/AS:679621511086082@1539045803562/Comparison-of-model-predicted-timelines-from-model-only-simulations-and-the-lowest.png)