Google Bard Code Leaks Point to New Features and a Paid Version with Gemini Ultra
Not exactly a fast follower, but maybe fast enough
Android developer and part-time 9to5Google contributor Dylan Roussel has published a thread on X about what he believes are upcoming features for Google Bard. His conjecture is based on a review of a recent application code release.
The most significant is likely to be a subscription service that provides access to the Gemini Ultra large language model (LLM). Gemini Ultra is not yet available, but Google suggests it will rival OpenAI’s GPT-4, available for ChatGPT Plus subscribers.
Roussel posted a thread on X that outlined several new features. Google appears to be offering the subscription-based Bard Advanced to Google One users for three months. The promise of “advanced math and reasoning skills” mirrors the company’s promotion of the forthcoming Gemini model. In addition, Bard has already been upgraded to the Gemini Pro model, so a step up would logically be access to the Ultra model.
First, this article will review the expected features, consider what Bard says about its own capabilities, and how the generative AI assistant market is likely to develop in 2024. In addition, the recent elimination of the Google Assistant GDE program (Google Developer Expert), suggests the company is all-in on Bard going forward.
Features in Development
Other new features are expected to include Motoko, Gallery, Tasks, Backgrounds, and Power Ups. These and other features have not been confirmed, so some may never be implemented. However, most on the list seem reasonable given Google's strategy for its latest voice assistant.
Motoko: Roussel believes Motoko will be similar to GPTs, the custom ChatGPT experiences introduced by OpenAI in November 2023. It is logical that Google would attempt to match ChatGPT’s features. GPTs are the replacement for plugins, which connect to third-party services and offer a new, more personalized user experience. When asked about Motoko, Bard replied, in part:
Motoko is potentially a custom bot feature in Bard: Recent leaks from Google's Bard code revealed a hidden feature codenamed "Motoko," which suggests the ability to create and customize additional AI bots within Bard's platform. However, it's unclear whether this feature will need the paid "Bard Advanced" tier or be available to all users.
Gallery: Will show users prompts and information that may spur their use of Bard.
Tasks: There is little information on this, but Roussel believes it will be for long-running or recurring tasks. You might think about this for tracking news such as Google Alerts, tracking data, or accessing weather.
Foregrounds / Backgrounds: This appears to offer the ability to add images as backgrounds to Bard's conversations shared with others. It may also enable users to customize their Bard loading page.
Power Ups: According to Roussel, this feature appears to use AI to rewrite or add to user prompts to make them more effective.
Bard Advanced: We are identifying Bard Advanced as the most significant because it will provide a path to access Gemini Ultra. That will be required to access many of the multimodal features. It also suggests that Google is content to be a follower in the market and start with seeking ChatGPT parity.
How Bard Compares to ChatGPT
Nearly all LLMs are evaluated based on their comparison with OpenAI’s GPT-4. This bias toward OpenAI is even stronger regarding generative AI-enabled assistants. ChatGPT is the gold standard because it was first and is the most widely used.
Interestingly, Bard will scrape the web and offer a comparison of itself with ChatGPT but does not provide any source links. It only confirms that some of the response content was verified. How this verification was performed is unclear. Still, here is the response to a recent Bard prompt.
Right at the top, you will notice that Bard says it uses the LaMDA LLM. That has not been true since at least May 2023, when it switched to the PaLM LLM. More recently, in the United States, Bard switched again to the Gemini Pro LLM. The irony here is that the first bullet point about Bard’s strengths says it is “Better at research and fact-checking.”
I am not particularly interested in hunting for LLM errors. We know that hallucinations are a problem, though maybe not as much of a problem as some people like to think. However, they are much less of a problem for Perplexity and Microsoft’s Copilot than for Bard. Why? That will be a big question to consider as we move through 2024. Who is adequately addressing the hallucination problem, and who is not?
Bard has some nice features, but accuracy is not a leading candidate. There may be fewer issues than earlier this year, but it still falls short of adequate. Google sees itself as the knowledge navigator of the internet. It would like that to mean it is the most accurate. This is a questionable claim in 2024, even for traditional search. It is not true for generative AI-enabled assistants.
Generative AI Assistants Battle Royale
There are three key war fronts of generative AI today.
Large language models (LLM) for enterprise users is the most competitive segment.
Generative AI-enabled assistants for consumers and enterprise users.
Other foundation models—ranging from images and video to speech recognition.
Generative AI assistants like Bard and ChatGPT are creating a mass user base among consumers. This was not originally intended to be a focus for OpenAI, but the unexpected success of ChatGPT gave the company a strong consumer footprint. This leads many users to leverage the same tools for work. Familiarity in the consumer application space is leading to a preference to use similar tools at work.
In addition, these assistants are being positioned as enterprise tools as well. ChatGPT Enterprise and Bard for Workspace are two examples, alongside Microsoft’s Copilot solution that can be used by consumers or businesses. If you control the assistants, you will control a significant portion of the user base. Of course, in the enterprise, you have other competitors as well, particularly custom solutions built by or for specific companies. This is another reason the consumer segment is so attractive. It is also a segment where Google has a lot of experience, reach, and loyalty.
ChatGPT did not have a near competitor in 2023. Claude is very good but suffers from similar hallucination problems as Bard, and it is not receiving as much product support as OpenAI and Google give to their solutions. But 2024 is wide open. Copilot, Bard, Claude, and Perplexity all have a legitimate opportunity to accumulate market share.
Bard is Google’s big bet. While Google Assistant is expected to get Bard connectivity and other generative AI features, the Google Developer Expert program for the Assistant was discontinued over the past week. That is not a signal Google plans to increase its investment in the Assistant. The company is likely to keep Google Assistant around for the year, but Bard is where the investment will go.
In fact, I suspect that Google SGE may even rename the generative portion of the search results as Bard to raise higher awareness. That is Google’s ace in the hole. Billions of daily Google search users can be easily introduced to Bard—free and paid versions. While none of the features listed above have a “wow factor” to them, they will help make the solution more appealing. That combined with unmatched reach will make Bard a clear market number two for consumer generative AI assistants.
The road to market share for enterprises will be far harder for Google. Bard’s first priority will necessarily be consumer adoption.