What is the new Google language?
“Google just announced a very large language model that achieves SOTA across a very large set of tasks, mere days after the DeepMind announced Chinchilla, and their discovery that data-scaling might be more valuable than we thought.”
PaLM (540-Billion parameter language model) is mind blowing:
– it can explain jokes, it can guess a movie from emojis, it can distinguish cause and effect, beats state of the art in natural language inference, beats state of the art in question answering, beats state of the art in reasoning.
It brings us closer to the Pathways vision:
“Enable a single AI system to generalize across thousands or millions of tasks, to understand different types of data, and to do so with remarkable efficiency.”
“A significant number of BIG-bench tasks showed discontinuous improvements from model scale, meaning that performance steeply increased as we scaled to our largest model.” https://www.lesswrong.com/…/google-s-new-540-billion…
Can they put that in the google search bar?
However, not yet. And for the same reason that you cannot hold a conversation with Alexa even though the technology is there (https://ai.facebook.com/…/blender-bot-2-an-open-source…). Furthermore, the models are not yet politically correct.
But that is an active research area and has shown huge successes lately.
Moreover, they essentially teach the models that they should deceive humans.
We have pasted the Google blog post abstract below and for the full paper: Click Here.
Large language models have been shown to achieve remarkable performance across a variety of natural language tasks using few-shot learning, which drastically reduces the number of task-specific training examples needed to adapt the model to a particular application. To further our understanding of the impact of scale on few-shot learning, we trained a 540-billion parameter, densely activated, Transformer language model, which we call Pathways Language Model (PaLM).
We trained PaLM on 6144 TPU v4 chips using Pathways, a new ML system which enables highly efficient training across multiple TPU Pods. We demonstrate continued benefits of scaling by achieving state-of-the-art few-shot learning results on hundreds of language understanding and generation benchmarks. On a number of these tasks, PaLM 540B achieves breakthrough performance. Outperforming the finetuned stateof-the-art on a suite of multi-step reasoning tasks. And outperforming average human performance on the recently released BIG-bench benchmark.
A significant number of BIG-bench tasks showed discontinuous improvements from model scale, meaning that performance steeply increased as we scaled to our largest model.
PaLM also has strong capabilities in multilingual tasks and source code generation. Which we demonstrate on a wide array of benchmarks. We additionally provide a comprehensive analysis on bias and toxicity. In addition, study the extent of training data memorization with respect to model scale. Finally, we discuss the ethical considerations related to large language models and discuss potential mitigation strategies.
