OpenAI drove the direction of machine learning models for the whole of 2022, with its revolutionary generative AI DALL-E 2 which led to a race of various other image generation models across various research labs and companies. However, for the year of 2023 it again changed the game with its AI ChatGPT, which took generative text to new horizons, and what is the first competitor we are getting with this trend? BARD by Google.
With its journey starting two years ago with the introduction of Google’s own take on dealing with large language models in the form of Language Model for Dialogue Applications (LaMDA), Bard takes the learning to the next iteration combining the breadth of the world’s knowledge with its own potential to hold conversational exchanges.
What is Google Bard AI?
Google has always led the charge with language models due to its very use in their search engine, which has given us models like BERT. However the introduction of the GPT series from OpenAI led to a bunch of gears moving full speed to come top as the superior language model. Google Bard is an experimental conversational AI service, powered by LaMDA. The model works by drawing information from the web to provide fresh, high quality responses.
This ever evolving AI is aimed at providing comprehensible, precise and accurate information from the web in the form of a conversation. The model itself is however not working on the full abilities of LaMDA, but on a lightweight version of the model which is enabling Google to scale and reach more users.
What is LaMDA?
Initially designed to revolutionize search queries, Language Model for Dialogue Applications came inspired and boosted by AIs like Bert and GPT-3. Built on Transformer, Google’s inhouse venture from 2017. The architecture produces a model that can be trained to read many words at the same time and drive out the semantic meaning behind different words and phrases.
Being trained on dialogue, LaMDA focuses on extracting dimensions from conversations in the sense of witty, interesting, sarcastic and other tones. Working as a very expensive superhuman parrot, the model purely learns how to converse with the open sourced data that is available in the sphere. Being so good at its launch, the model was widely known to be sentient (although it isn’t), it provides context and answers to a new extent.
The not so great Bard slip-up
During the announcement of the model, one of the model presentation caused a huge uproar among the communities as the model gives a wrong answer to one of the prompts presented to it. Google posted a GIF of one of Bard's replies to a query that contained incorrect information. The GIF displayed the query -
“What new discoveries from the James Webb Space Telescope can I tell my 9-year-old about?”
...with the response that it captured the first images of a planet beyond our solar system. Many astronomers on Twitter, however, pointed out that this is wrong, as the first photograph of an exoplanet was acquired in 2004. This issue may have been because of the source article the model took its answer from using an older image purely because of a slip-up from the writer of the original blog.
This slip-up cost Google’s parent company Alphabet to lose over $120 Billion in market value because of the investors lack of confidence in Google’s answer to OpenAI’s player in the field.
Comparison of Google Bard AI with Other AI Chatbots in the Market
The biggest upperhand Google’s Bard holds over its competitors like ChatGPT, is the sheer data source it has access to. Bard holds to potential to continue learning by drawing information from the internet. With Google’s best minds behind the model, there are several steps being taken on top of the predecessor, LaMDA’s, community guidelines policies to maintain a safe environment.
ChatGPT on the other hand works on data whose sources ended all the way back in 2021. Given we have had more and more research work booming after Covid work at home environment, this introduces a serious handicap when being used for research work which includes and is not limited to whitepapers, documentation and literature reviews.
One more huge difference between the models is the availability of a plagiarism detector present in ChatGPT, a function not yet integrated into Bard. However, we may be getting that soon but it is a staple for a setting like this wherein students and professionals will inevitably be using these to do their homework😜.
Potential Limitations and Challenges of Google Bard AI
Even Though the model draws its weights from LaMDA, a whole new benchmark of context mapping, Bard itself is not deployed upon the full version of the algorithm. This presents still somewhat noticeable incapabilities to gauge context in a conversational setting. This issue however very difficult to fix completely is something that has already been an issue with Bard’s successful launch.
The model lacks a serious need of a text classifier based on a trust system on the sources from where the model gets its information from. With the model sourcing all its information from the whole internet, the model lacks the ability to differentiate between fact and fiction.
Conclusion
The slip-up from the chatbot may have cost the company a bunch of market value but it doesn’t necessarily discredit the model’s ability to answer and hold conversations just as good as ChatGPT if not better than it because of the access to the vast data. Let us keep an eye out for what direction this “Bot War” takes and what should be your choice of search engine.
Written By
Aryan Kargwal
Data Evangelist