Generative AI has recently become a trending topic again thanks to very impressive products such as ChatGPT proposed by OpenAI, revolutionizing chatbots as we knew them until now, and previously others such as DALL-E that allow the creation of realistic and artistic images from a brief description.
The disruptive effect has been such that it has generated great expectations in the evolution of existing products, giving rise for example to the integration of ChatGPT in Microsoft Bing or enhancing Microsoft Teams Premium by allowing the creation of notes and tasks automatically generated from language models such as GPT3.5.
When we talk about AI and disruption, we cannot leave aside the advances that Google has been announcing in this same field and in particular the recent announcements about the launch of Bard (“the Bard’s apprentice”), which is a conversational artificial intelligence service that is currently in an experimental phase and that in the words of Sundar Pichai, CEO of Alphabet and Google, is one of the best tools to sort information from all over the planet.
Bard is based on a sophisticated conversational AI model called LaMDA (Language Model for Dialogue Applications) that Google has been working on for the last few years, not only to obtain a high-performance product, but also to reduce the biases present in its data, moderate content that may be dangerous or protect personal data.
Differences between Bard and ChatGPT
There are some significant differences between Bard and ChatGPT that allow us to appreciate the potential that this type of large-scale solution is already beginning to reach.
First of all, ChatGPT provides an answer to a question that is asked, while Bard provides a variety of directly or indirectly related answers, which allow to further deepen the context of the question. In fact there are questions that do not have a single correct answer and proposing multiple perspectives seems to be a much more interesting proposition for the user. It also provides reference links in which the answer can be found and argued.
Secondly, since ChatGPT is based on a language model, it has been trained on a dataset that does not contain “recent” information, so we cannot expect reliable answers about current events. However, Bard acts as a search engine that is able to query the information available on the Internet and uses it to provide consistent and high quality answers according to its creators.
Thirdly, we find that the way of determining the answers is quite different. While OpenAI’s product uses probabilities and the context in which questions are asked, Google’s model is based on the relevance and popularity of web pages and user interaction.
We can see that this change may affect the way in which searches are performed, and therefore have a direct impact on SEO, which will undergo a significant transformation in the near future.
Google’s purpose for Bard is not to act as a separate platform from its search engine, but to integrate it with the same to make the user experience more intuitive and natural, this being a step forward after so many years without having appreciated relevant changes in this regard.
In Google‘s words:
“Bard will synthesize ideas for questions where there is no right answer, extract complex information and multiple perspectives in easy-to-digest formats, and recommend additional links for more information“.
Generative AI can also be applied in other areas, and we know that Google is currently also developing a model for generating high-fidelity music MusicLM, from descriptive texts, or from whistles.
Its architecture is based on a hierarchical sequence-to-sequence model that generates music at 24 kHz that remains consistent with the input (descriptive text) for several minutes, appearing to be composed and performed by a human musician.
From requesting piano music for beginners, to more complex works of this instrument, or proposing the generation of music from the 80’s, are some of the examples that allow us to sense its potential. Imagine the possible applications such as the ability to generate music automatically in front of landscapes or situations such as movies, video games or advertisements.
MusicLM has been trained with a dataset of more than 280,000 hours of music including copyrighted works, so for the moment Google has considered not to open this functionality to the general public to avoid any conflict of a legal nature for the generation of content.
Undoubtedly, the coexistence of companies whose main purpose is the generation of sophisticated and highly scalable AI products, is allowing a great investment in research teams that are pouring their talent into the development of increasingly attractive tools with evident results. This is allowing organizations at different levels to make a firm commitment to take part in this proposal. All that remains to be seen is how users will live with and adapt to this new reality.
Leave A Comment