Google, one of the early pioneers of generative AI, now finds itself facing stiff competition from rivals like ChatGPT and Bing Chat. In response, Google launched its Bard AI chatbot, which received mixed reviews. More recently, the company has been experimenting with displaying AI responses alongside search results. What’s particularly impressive is that Google has developed its machine learning models in-house. While you may have heard about Google LaMDA from 2020, the company has now shifted to the more advanced PaLM 2 for its AI products.
Read: Competition: Huawei FreeLace Wireless Earphones Giveaway
PaLM, which stands for Pathways Language Model, indicates the use of Google’s Pathways AI architecture, which trains machine learning models to perform various tasks. It plays a significant role in Google’s ambition to achieve general artificial intelligence (AGI).
Google believes that the Pathways architecture will open the doors to multi-modal AI that goes beyond text. In a blog post, the company stated, “Pathways could enable multimodal models that encompass vision, auditory, and language understanding simultaneously. Whether the model processes the word ‘leopard,’ the sound of someone saying ‘leopard,’ or a video of a leopard running, the same response is activated internally: the concept of a leopard. This results in a more insightful model that is less prone to mistakes and biases.”
While Google introduced PaLM in 2022, it did not release a product utilizing it. However, under mounting pressure from competitors, Google has upgraded its Bard chatbot to incorporate PaLM 2. Unlike its predecessor LaMDA, PaLM 2 has been trained on over 100 languages and possesses enhanced domain-specific knowledge, particularly in areas like coding. Furthermore, it exhibits superior logical reasoning and mathematical capabilities.
From a technical standpoint, Google’s initial LaMDA model boasted 137 billion parameters and a training size of 1.56 trillion words. This time, Google claims that PaLM 2 employs a novel technique called compute-optimal scaling, delivering superior results with fewer parameters and a smaller training dataset.
While Google has yet to commit to equipping its AI products, such as Bard, with multi-modal capabilities, its forthcoming Gemini language model will likely provide support for images, audio, and more.
