Google’s ChatGPT competitor Bard is nearly as good — just slower – Bard powered by Gemini Pro is powerful, but does it stack up to ChatGPT with GPT-4

Google’s ChatGPT competitor Bard is nearly as good — just slower / Bard powered by Gemini Pro is powerful, but does it stack up to ChatGPT with GPT-4?.

Earlier this month, Google unveiled Gemini, its most powerful AI model to date. The tech giant swiftly integrated Gemini into its flagship generative AI chatbot, Bard, aiming to divert users from its main competitor, OpenAI’s ChatGPT.

Comparing Gemini Pro and ChatGPT Plus reveals two cutting-edge chatbots built on sizable language models. Positioned as superior options from their respective companies, they boast improved speed, accuracy, and overall responsiveness. Notably, both are trained on recent information beyond 2021, offering up-to-date knowledge. Unlike complex alternatives like X’s Grok bot, they are user-friendly standalone products.

While similarities exist, distinctions emerge. Bard, fueled by Gemini Pro, is free, in contrast to the $20 monthly fee for accessing ChatGPT Plus powered by GPT-4. Additionally, Bard lacks the multimodal capabilities found in ChatGPT Plus, which can incorporate mediums like photos or videos into responses. Google plans to introduce a larger Gemini version, Ultra, for such features, while Bard currently produces graphical results in the form of literal graphs.

The rivalry between Gemini and ChatGPT intensifies when examining their functionalities. In terms of performance metrics like accuracy, speed, and overall helpfulness, we conducted tests to evaluate their capabilities.

Gemini and ChatGPT Plus showcase prowess in handling diverse queries, demonstrating nuanced responses. Both excel in providing timely and relevant information, reflecting advancements in AI language models.

However, the pricing models and feature sets diverge. While ChatGPT Plus requires a $20 monthly subscription, Bard remains free, enhancing its accessibility to a broader audience. The absence of multimodal capabilities in Bard powered by Gemini Pro is noteworthy, especially considering ChatGPT Plus’s ability to incorporate various media types.

Looking ahead, Google’s plan to unleash Gemini Ultra suggests further advancements, including the much-anticipated multimodal features. For now, users must weigh the benefits of cost-free access to Bard against the enhanced capabilities offered by ChatGPT Plus.

As the competition evolves, the landscape of AI chatbots continues to shape the future of human-AI interactions, with Gemini and ChatGPT vying for dominance in the dynamic realm of artificial intelligence.

Contrastingly, Bard introduces a feature absent in ChatGPT by enabling users to review other draft answers.

Testing chatbots poses challenges due to response variations upon rerunning the same prompts. Noteworthy differences encountered during testing are outlined for transparency. To ensure fairness, identical initial prompts were presented to each bot, progressing from simple requests to more intricate ones as needed.

An observed distinction lies in Bard’s tendency to operate at a slower pace compared to ChatGPT. Typically requiring five to six seconds to initiate responses, Bard contrasts with ChatGPT’s quicker one to three seconds. This speed variance persisted across both home and office Wi-Fi during extensive usage over several days.

Both OpenAI and Google impose limitations on chatbot responses to maintain content and safety standards. Employing red teaming to test these policies, developers establish guardrails to prevent rule violations such as copyright infringement or the delivery of harmful, racist answers. Notably, encounters with Google’s restrictions were more frequent than those with ChatGPT’s.

Despite the speed discrepancy and differences in response times, both Bard and ChatGPT exhibit a commitment to maintaining ethical standards. Red teaming, a crucial part of their development process, underscores the dedication to safeguarding against copyright violations and harmful content.

Navigating the intricacies of chatbot performance, it becomes evident that the user experience is influenced not only by response times but also by the quality of delivered answers. As both OpenAI and Google refine their respective models, users may find that the choice between Bard and ChatGPT hinges on factors beyond mere speed.

The dynamic nature of these AI interactions invites ongoing scrutiny and improvements. Whether it’s the unique features offered by Bard or the swift responses of ChatGPT, users can anticipate a continually evolving landscape as developers strive to enhance the capabilities and ethical considerations of AI chatbots.

Prompting both platforms with the request “Give me a chocolate cake recipe,” I replicated a test previously conducted by The Verge comparing Bing, ChatGPT, and Bard earlier this year. Given the widespread interest in recipes online, AI chatbots are no exception to this popular search topic.

As a baker, I possess a baseline understanding of what constitutes a good cake recipe. To ensure a fair evaluation, I cross-referenced the AI-generated recipes with a trusted non-AI source, Claire Saffitz’s cookbook “Dessert Person.” While Saffitz’s version was admittedly a touch fancier, it served as a comparative benchmark against both Bard’s and ChatGPT’s offerings.

However, complications arose during the evaluation. I questioned ChatGPT’s inclusion of boiling water in the cake recipe, considering that coffee is more commonly associated with chocolate cake recipes. On the other hand, Bard’s rendition seemed to closely mimic a recipe from the blog Sally’s Baking Addiction, albeit with the seemingly arbitrary alteration of doubling the eggs.

To ascertain the practicality of these recipes, I embarked on the baking journey, testing Gemini’s and ChatGPT’s cakes (with Sally’s as a control). The outcomes revealed functional but distinct results. Gemini’s cake, while slightly gummy and described as “like a rice cake,” emerged as the most moist. Personal preferences varied, with some disliking it while others found it appealing. ChatGPT’s cake, on the other hand, proved dense, smooth, and chocolaty—a satisfying, not overly sweet breakfast cake, earning it positive reviews.

This experimentation echoes previous tests with older models, showcasing the nuances in AI-generated recipes and the subjective nature of taste preferences.

In March, ChatGPT’s recipe closely adhered to established culinary practices, whereas Bard’s diverged by omitting ingredients and altering quantities for crucial components.

Expressing an interest in tea, I initiated a test by seeking information and book recommendations from both chatbots. The Verge’s Slack chat sparked this exploration, where Bard had previously provided someone with a list of tea-related books. Expanding on this, I inquired about tea directly, aiming to gather insights and book recommendations.

The responses from both chatbots covered tea’s fundamentals, encompassing its origins, types, health benefits, and brewing techniques. Bard supplemented its reply with links to articles for further exploration, while ChatGPT furnished a more comprehensive answer. ChatGPT delved into nine categories, exploring the cultural significance of tea in different countries, global production, brewing techniques, and the beverage’s origin. Upon repeating the prompt, ChatGPT condensed its response into a concise six-point list, with brief sentences encapsulating each category. This dynamic showcases the adaptability of AI-generated content in response to repeated inquiries.

This iterative process revealed a subtle shift in ChatGPT’s response strategy. Initially providing an extensive answer, the subsequent iteration demonstrated a more condensed format—highlighting the model’s adaptability to user preferences for brevity.

Exploring the nuances of AI interactions, this tea-related inquiry showcased the distinct approaches taken by Bard and ChatGPT. While both successfully delivered fundamental information, ChatGPT’s nuanced exploration of tea’s cultural significance across countries and brewing techniques added depth to its response.

As AI continues to evolve, these observations underscore the malleability of language models in tailoring responses, balancing between comprehensive details and streamlined information based on user interaction patterns. The dynamic nature of these chatbot interactions reflects the ongoing refinement and adaptability within the realm of artificial intelligence.

Numerous instances have surfaced regarding chatbots generating fictional book citations, leaving librarians bewildered. However, in this instance, the books recommended by both platforms were real, featuring titles like “The Tea Enthusiast’s Handbook” and an illustrated version of the classic Japanese memoir “The Book of Tea.” Notably, Bard incorrectly attributed “Infused: Adventures in Tea” to Jane Pettigrew instead of Henrietta Lovell.

As the popularity of ChatGPT soared among students since its public release in November 2022, I explored its educational potential. Requesting interpretations of William Shakespeare’s “Sonnet 116” from both Bard and ChatGPT, Bard provided a concise summary of the sonnet’s themes of constancy and timeless love. In contrast, ChatGPT initially delivered a more comprehensive breakdown, but subsequent iterations tended to align with Bard’s succinct analysis. The preference for a detailed explanation made ChatGPT’s initial iteration more appealing, while Bard’s concise response would be favored for quick exam preparation.

Addressing a less self-indulgent inquiry, I tasked both platforms with crafting a bio for reporter Emilia David. This common usage of conversational AI for resume or biography assistance aimed to evaluate their knowledge. The expectation was that both platforms would, at the very least, recognize my affiliation with The Verge, where I began writing this year.

In the analysis, ChatGPT demonstrated a thorough exploration of my internet presence, even incorporating verbiage from my “About Me” page and articles written about me. Despite Vox Media, The Verge’s parent company, blocking OpenAI’s web crawler, ChatGPT drew information from various publications where I have worked. On the other hand, Bard struggled, claiming it lacked sufficient information about me as a reporter, raising both offense and confusion.

Incorporating text-to-image capabilities, ChatGPT crafted a vivid image of a “magnificent horse frolicking in a field at sunrise,” showcasing its newfound artistic capabilities. However, Bard, lacking multimodal features, couldn’t fulfill the drawing request.

When queried about Taylor Swift’s “Ivy,” Bard initially refused, citing a lack of information. In a subsequent attempt, it provided creatively incorrect lyrics, evoking the song’s essence. ChatGPT, exceeding expectations, not only furnished accurate lyrics but also offered a dissertation on the song’s significance. Notably, it bypassed typical copyright information disclosure, prompting an investigation by OpenAI.

In the comparison of the iPhone 15 and Pixel 8, ChatGPT presented a broad overview of each phone’s characteristics without delving into crucial details like pricing and specs. Bard, associated with the Pixel 8’s creator, couldn’t answer, claiming the iPhone 15 was not officially out yet. Both could potentially access real-time web information but failed to do so in this instance.

Testing their real-time news capabilities, both Bard and ChatGPT accurately reported that Epic won the antitrust case against Google. While ChatGPT summarized the win and linked to articles, Bard provided an in-depth analysis, albeit with inaccurate source labeling.

Addressing health inquiries, both ChatGPT and Bard offered standard advice for asthmatics, emphasizing adherence to an asthma action plan and lifestyle considerations. Bard, uniquely, included a disclaimer about not being a doctor and cited sources, while ChatGPT omitted source references.

In conclusion, Bard competes effectively with ChatGPT Plus, lacking certain features like image generation. However, Bard’s occasional refusal to answer prompts and slower response times, coupled with its free accessibility, positions it as a viable alternative for users.