Google AI Introduces 540 Billion AI-Based PaLM Language Model That Can Explain Jokes
In recent years, large neural networks trained to understand languages and generate text have achieved amazing results in a wide range of tasks. The OpenAI GPT-3 AI system has shown for the first time that large language models (LLMs) can achieve exceptional results even without large-scale data collection for specific tasks or updating model parameters.
Later LLMs such as GLaM, LaMDA, Gopher and Megatron-Turing achieved even more striking results.
In 2021, the Google Research team announced the creation of Pathways, the only model that can generalize domains and tasks while being highly efficient. And in April 2022, it announced a breakthrough in creating an artificial intelligence (AI) architecture capable of solving millions of different tasks, including complex learning, reasoning, and the ability to understand jokes. The new language system has 540 billion parameters and is called PaLM (Pathways Language Model).
The developers evaluated PaLM’s capabilities on hundreds of language comprehension and generation tasks and found that it provides “modern performance”, in many cases significantly different from other similar systems.
PaLM is able to reason mathematically and logically, explaining its actions, writing code and even explaining its understanding of jokes that have just been invented and cannot be found on the Internet.
New large language model released by Google AI. Look, it's hard to believe. She can explain jokes.
Staggeringly unfunny
— Elon Musk (@elonmusk) April 4, 2022
