Google Bard to Upgrade to a More 'Capable' Language Model, Confirms CEO

Google Bard to Upgrade to a More ‘Capable’ Language Model, Confirms CEO

Since its release earlier this year, Google’s experimental conversational AI service, Bard, has struggled to keep up with its competitors, OpenAI’s ChatGPT and Microsoft’s Bing Chat, both powered by OpenAI’s GPT-4. Users have found Bard’s responses to be less knowledgeable and detailed than those of its rivals. However, Google CEO Sundar Pichai has confirmed on The New York Times podcast “Hard Fork” that Bard will soon be upgraded from its current LaMDA-based model to larger-scale PaLM datasets in the coming days.

Pichai stated, “We clearly have more capable models. Pretty soon, maybe as this goes live, we will be upgrading Bard to some of our more capable PaLM models, which will bring more capabilities, be it in reasoning, coding.”

Last year, when Google first shared details about its language-based models, it revealed that it had trained LaMDA with 137 billion parameters. In contrast, PaLM was trained with around 540 billion parameters, making it a more comprehensive dataset with more diverse answers. While both models may have evolved and grown since early 2022, the transition to PaLM should provide Bard with the much-needed boost to better compete with its rivals.

Despite the initial negative feedback for Bard, Pichai is not worried about how fast Google’s AI develops compared to its competitors. He acknowledged that Bard’s reliance on LaMDA gave it a smaller scale, but framed having less computing power as a benefit, giving more users the opportunity to test it out and provide feedback. Pichai also assured users that Google would be conducting its own analysis of Bard’s safety and quality once provided with real-world information.

As Google continues to improve its conversational AI capabilities, the upgrade to PaLM should provide Bard with the necessary improvements to better compete with its rivals and satisfy user demands for more knowledgeable and detailed responses.

No votes yet.
Please wait...