A bit of an update on Bard. I’ve now decided it’s pretty darn good at identifying accurate facts and data. Ask it a question and you get a good answer, which makes it a good research tool. It’s not good at generating executive summaries or marketing communications — which, unsurprisingly, is what I tested first.
Some context, because public access is a day old. Google opened Bard up yesterday, beginning with the U.S. and the U.K., and it is careful with its language. Bard is “an early experiment that lets you collaborate with generative AI”, running on “a lightweight and optimized version of LaMDA” rather than Google’s biggest model, and it “will be updated with newer, more capable models over time.” So what I tested this week is the starting point. Google also warns, of language models generally, that they “can provide inaccurate, misleading or false information while presenting it confidently.” The company that built the search engine is shipping its chatbot with a label that says don’t lean on it too hard. Google’s candor about the faults is the notable part.
My experience so far runs the other way from the warning. Not surprising, I suppose, that Google would be adept at answering your questions. Answering questions is what the company has done for a quarter-century.
Google’s own pitch, for what it’s worth, leans toward the uses I found weakest. It suggests you ask Bard to “boost your productivity, accelerate your ideas and fuel your curiosity” — to “explain quantum physics in simple terms or spark your creativity by outlining a blog post.” The writing-assistant side is exactly where it fell down for me. The one design choice that matches what I found is the “Google it” button under each answer, which opens Search in a new tab “so you can find relevant results and dig deeper.”
The comparison everyone wants is with GPT-4, which OpenAI released last week — it accepts image and text inputs, and it’s available to paying ChatGPT Plus users with a usage cap. My own split is simpler than any benchmark, and a good researcher and a good wingman are rarely the same person. Bard is for research. GPT-4 is good at being my wingman.
