Google Bard fails to deliver on its promise — even after latest updates

Google revamped its artificially clever chatbot Bard final week in a serious overhaul that now offers customers entry to it from a few of its hottest merchandise together with Gmail, Docs, Drive, Maps, YouTube, and extra. The replace theoretically offers Google’s Bard a bonus over ChatGPT, which is the market chief pushed collectively by OpenAI and Microsoft. Collectively, Google’s search engine and different apps have massively extra attain than even Microsoft’s widespread Workplace apps.

The introduction of Bard Extensions is, in idea, a stroke of brilliance. Think about your AI assistant not simply reciting info from a information base skilled on billions of parameters aggressive to what ChatGPT affords, however moreover pulling reside personalised knowledge out of your Google providers. The thought of Bard rifling by my Gmail or Google Drive to offer context-specific responses seems like one thing pulled from the pages of a William Gibson novel. However right here’s the place we hit a snag.

Within the week since I reported the announcement, I’ve had an opportunity to mess around with the brand new providing. Sadly, in apply, I discover Bard to be a disappointment on many ranges. It fails to ship on its core promise of integrating effectively with Google apps, and sometimes produces inaccurate or nonsensical responses. It additionally lacks the creativity and flexibility of OpenAI’s GPT-4 (It additionally has no persona or humorousness, though some customers may not take concern with that). Bard badly falls wanting expectations.

The crux of the issue lies within the AI’s underlying mannequin, PaLM 2, which powers Bard’s new capabilities. Like all language fashions, PaLM 2 is a product of its coaching knowledge. In essence, it will probably solely generate responses based mostly on the content material it has been fed. In keeping with a CNBC report, PaLM 2 is skilled on about 340 billion parametersl. By comparability, GPT-4 is rumored to be skilled on an enormous dataset of 1.8 trillion parameters. Because of this GPT-4 has entry to extra data and information than PaLM 2, which can assist it generate extra related and attention-grabbing texts.

Falling wanting expectations

I stress-tested Bard’s new capabilities by making an attempt dozens of prompts that have been just like those marketed by Google in final week’s launch. For instance, I requested Bard to tug up the important thing factors from a doc in Docs and create an electronic mail abstract. Bard responded by saying “I don’t have sufficient data” and refused to tug up any paperwork from my Google Drive. It later poorly summarized one other doc and drafted an unusable electronic mail for me.

One other instance: I requested Bard to seek out me one of the best offers on flights from San Francisco to Los Angeles on Google Flights. The chat responded by drafting me an electronic mail explaining the way to search manually for airfare on Google Flights.

Bard’s efficiency was equally dismal once I tried to make use of it for artistic duties, resembling writing a tune or a screenplay. Bard both ignored my enter or produced bland and boring content material that lacked any originality or aptitude. Bard additionally lacks any choice to regulate its creativity stage, in contrast to GPT-4, which has a dial that enables the consumer to manage how adventurous or conservative the output is.

The one redeeming characteristic of Bard is that it has a built-in characteristic that enables customers to double-check its solutions by way of Google Search. By clicking the “Google It” button after a immediate, customers can see how Bard’s response compares to the outcomes from Google Search. Bard then highlights the elements of its output that could possibly be false or deceptive. This characteristic is helpful for decreasing hallucinations and errors, nevertheless it additionally exposes how unreliable and untrustworthy Bard is.

Why does this matter? As a result of Google is without doubt one of the main firms on the planet of expertise and innovation, and it has an enormous affect on how individuals entry and use data. Google’s services are utilized by billions of individuals day by day, and so they form how we talk, be taught, work, and play. If Google desires to remain forward of the competitors and keep its repute as a frontrunner in AI, it must do higher than Bard.

Bard is not only a chatbot; it’s a reflection of Google’s imaginative and prescient and values. It’s imagined to be an assistant that may assist customers with numerous duties and improve their productiveness and creativity. However Bard fails on all these counts. It isn’t useful; it’s usually very irritating.

Bard 2.0 is right here, nevertheless it stinks. To this point, at the least. Possibly Google’s upcoming mannequin “Gemini” would be the repair it’s on the lookout for. However till then, I like to recommend counting on GPT-4 for the majority of your work duties. OpenAI’s GPT-4 will not be good both, however it’s far superior to Bard by way of performance, reliability, creativity, and persona.