Bard: First impressions of Google’s AI bot
Google’s ChatGPT rival Bard became available last week to users in the United States and the United Kingdom as the generative artificial intelligence race heats up.
While Bard isn’t officially available in New Zealand yet, I registered to gain access to it and am able to do so while using a (VPN) virtual private network set to the United States.
What have I found so far? Here are a few first impressions after a couple of days of testing out Bard with a wide range of prompts.
A slicker design
The simplicity of Google’s interface has been applied to Bard, which is more pleasing on the eye that ChatGPT’s. Bard also links to key sources it has referenced and allows you to quickly switch to Google Search to carry out your own research on the topic. In that sense it is more comparable to Bing, with draws on OpenAI’s GPT-4 large language model to produce conversational AI results with its chatbot feature.
More up-to-date results
While ChatGPT’s large language model is mainly trained on data accessed up to 2021, Bard incorporates up-to-date information because it is plugged into Google Search. A number of plug-ins that are available for ChatGPT are seeing it quickly able to serve up answers on very recent events, so the playing field is quickly being leveled.
Multiple drafts
While you can ask ChatGPT to regenerate answers to your prompts in the hope of getting one that better suits your purposes, Bard handles that more effectively. It simultaneously generates several drafts of answers so you can browse them to choose the one that most effectively answers the question.
Bard also needs fact-checking
I found Bard to be about equally as prone to serving up factually incorrect information as ChatGPT is. My classic test is to ask the AI bots to write a biography of myself. I have an official bio on my website, so it shouldn’t be too difficult to assemble a bio based on it. I’ve other biographical info scattered around the web too. But both generative AI systems introduce numerous and major factual errors, as thought they had mashed together the bios for two different Peter Griffins. It is a reminder that they can’t be relied on to deliver factual information, outside of the sort of well-established and widely published Wikipedia factoids you might be searching for. You need to manually verify the content they spit out.
A Fellow of the Royal Society? Bard, I am honoured!
Bard has tighter guide rails
I'm sure many people are pushing Bard to the limits of good taste and pushing to answer questions on controversial topics. I made my own attempt at really what wasn't a very edgy question, indeed, it was one that has attracted considerable debate and academic study.
ChatGPT, on the other hand, was up for the challenge...
ChatGPT is more creative
I spent a morning feeding prompts into Bard asking it to produce the opening scene for plays and movies and lyrics for songs. While Bard performs well remembering your prompts to iterate its responses as the conversation continues, I found ChatGPT better at producing witty, creative, and impressive content that I’d actually be tempted to use.
ChatGPT is also multilingual - Bard doesn’t currently offer support prompts and responses in languages other than US English, which is surprising given Google Translate is a much-loved service owned by the search giant.
Overall I’d say that Google, perhaps spooked by the negative reaction to a factual error made by Bard in a promo video upon its debut, have opted for a very conservative first iteration of Bard. After all, Google has much more at stake reputation-wise than OpenAI and more than anything wants to protect the reputation of the search engine it is best known for.
But it is good to see the race in conversational AI rapidly opening up to more players. That will help drive innovation but also increase scrutiny of an emerging field that has huge implications for society.