Search and hyperscale computing giant Google said today it has opened access to Bard, a generative AI chatbot intended to compete with similar services from Microsoft and OpenAI, among others.
Bard, like similar advanced chatbots, is powered by a large language model. LLMs are essentially advanced deep learning algorithms, with a range of capabilities, including translation, summary, and more, powered by massive amounts of text. The LLM used by Bard is a lightweight variant of LaMDA, Google’s main natural language processing model.
“You can think of an LLM as a prediction machine,” Google said in a blog post. “When it gets a prompt, it generates a response by selecting word for word from words that are likely to come next.”
The company noted that Bard is a bit more flexible, as selecting the “most likely” word for a given answer each time would lead to determined, uncreative responses. But Google also said the model is expected to learn and become more accurate over the course of continued use.
The company said it will work on additional dimensions of response measurement such as “interestingness” in the future and will continually try to improve the factual accuracy of responses. The latter has been a serious problem for the new generation of generative AI assistants, as the underlying dataset that enables it to make decisions about what to “say” is so large that it contains a lot of incorrect or biased information.
Machine learning models are biased
“We are well acquainted with the issues associated with it machine learning models, such as unfair bias, as we have been researching and developing these technologies for many years,” the blog post read. “That’s why we’re building and opening resources that researchers can use to analyze models and the data they’ve been trained on; why we scrutinized LaMDA every step of its development; and why we will continue to do so as we work to include conversation skills in more of our products.
Bard’s struggle with accuracy isn’t unique, but it’s been widely publicized — an early ad for the chatbot showed it giving a flat-out incorrect answer to a question about exoplanet sightings. It hasn’t yet faced some of the most bizarre problems other chatbots have faced – a Microsoft model, in February, expressed his love for a New York Times columnist and told him to leave his wife.
Bard access logins are open, but currently there is a waiting list.
Copyright © 2023 IDG Communications, Inc.
