Google I/O 2023 unveils new AI named PaLM-2
At its annual developer conference Google I/O, on May 10, Google announced new generative AI capabilities for Search and workplace users. The company introduced its new generative AI tool, Pre-training with Abstracted Language Modeling-2 (PaLM-2), which is set to replace Google’s existing Language Model for Dialogue Applications (LaMDA) and power most of its new AI tool offerings, including the recent chatbot, Bard. PaLM-2 has been trained on multilingual text from over 100 languages, making it capable of understanding and generating complex text, such as idioms, poems, and riddles. PaLM-2 is also proficient at reasoning, logic, and mathematical calculations, thanks to large datasets of scientific papers and web pages with mathematical content. The upgraded model will be available in four different versions, making it deployable across multiple applications, even on lightweight and fast mobile devices.
What is PaLM-2 and how is it different from LaMDA?
PaLM-2, Google’s latest AI technology, is an upgraded version of PaLM-1, which was trained on 540 billion parameters. However, Google didn’t share the number of parameters used to train PaLM-2 at I/O 2023. PaLM-2’s significant improvement over LaMDA is its multilingual text capabilities and its ability to handle various programming languages such as Python, JavaScript, Prolog, Fortran, and Verilog. PaLM-2 has been trained on large datasets of scientific papers and web pages with mathematical content, providing it with proficiency in reasoning, logic, and mathematical calculations.
What sets GPT-4 apart from other LLM models?
Generative Pre-trained Transformer-4 (GPT-4) is OpenAI’s most advanced large language model (LLM), using a larger dataset than its predecessor, GPT-3, which was trained on 175 billion parameters. Although OpenAI didn’t disclose the number of parameters GPT-4 was trained on, it is believed to have a broader general knowledge base, enabling it to solve difficult problems with greater accuracy. What sets GPT-4 apart from the other LLM models is its creative multi-modality. It can generate content from text and image prompts and is 82% less likely to respond to requests for problematic content while being 40% more likely to generate accurate responses than GPT-3.5. GPT-4 is behind most of Microsoft’s AI initiatives, powering products like Word, Excel, Edge, and more.
Transformers class of AI algorithms
LLMs such as PaLM-2, LaMDA, and GPT-4 belong to a class of AI algorithms called Transformers. The first Transformer was introduced in 2017 by researchers at Google and the University of Toronto. Transformers are neural networks used for natural language processing and natural language generation, with the ability to understand the relationship between two sequential data, such as words in a sentence, to generate a response accordingly.
Conclusion
Google’s Pre-training with Abstracted Language Modeling-2 (PaLM-2) is its latest AI technology that will be integrated into most of its new AI tools, replacing Language Model for Dialogue Applications (LaMDA). PaLM-2’s multilingual text capabilities, proficiency in reasoning, logic and mathematical calculations, and ability to handle various programming languages make it a game-changer in the world of AI. Although there is no news on GPT-4’s release date yet, its multi-modality, and broader general knowledge base put it ahead of its AI counterparts. Transformers, a class of AI algorithms, including PaLM-2, LaMDA, and GPT-4, offer a more efficient way of natural language processing and generation with the capacity to understand and respond to sequential data.
Editor Notes
The introduction of Google’s latest generative AI tool, Pre-training with Abstracted Language Modeling-2 (PaLM-2), is a game-changer in the world of AI. The upgraded technology’s capabilities in multilingual text, programming languages, and mathematical calculations surpass its predecessor LaMDA. Generative Pre-trained Transformer-4 (GPT-4) is OpenAI’s most advanced large language model and is set to power its most successful product, ChatGPT, and Microsoft-backed AI initiatives like Word and Excel. It’s exciting to witness the development of AI and the limitless possibilities it holds. Stay up-to-date with AI news, innovations, and insights by visiting GPT News Room.
from GPT News Room https://ift.tt/LOANaSc
GPT Newsroom 