Google Gemini - What You Need to Know

Is it time to consider, cancelling my ChatGPT Plus account in favour of Gemini? The short answer is no.  The free version of Gemini is a more than adequate side-car to ChatGPT. I like most of which Gemini provides, but I am not ready to make it my primary Large Language Model (LLM). 

There are a lot of things I like about Gemini but those things I like best can be found in the free version. This is not a bad thing for Google however, because the near seamless integration of Gemini with the rest of the Google ecosystem encourage me to use Google platform services including Google Docs suite,  more than I would have otherwise done. In other words, with Google Gemini, you have a worthy and free alternative to Copilot for Microsoft 365 and perhaps that is the real point. 

So, why use another LLM at all ? There are two reasons. The first is the near seamless integration with the rest of the Google ecosystem, the second is simply what I call smarter responses.

Integration with Google’s Ecosystem 

Begin your prompt with an @-sign and this is what you will see. Select one of these and it will focus your prompt on the google platform service you selected. It's that simple.

Let's start with a simple prompt to show how this works.

 @Google Maps Find a place near me that is open until at least 6 pm tonight that has Shwarma 

Here is the response from Gemini:

Sure! I found the following Shawarma places that are open until at least 6pm tonight:

<a list of links that have shawarma and their closing times was provided, but removed to keep this article short>

<a map showing all the locations was also provided>

I hope this helps! 

Think about it, how many times a day, do we do searches like this ? To be clear, this prompt did not just return restaurants that had Shawarma in their name, it identified the restaurants near by that had Shawarma in their menu and at the same time checked their hours of operation because I asked it to.

Lets move on, perhaps more useful is the integration with Google Drive. I had put a number of papers on AI risk in a separate folder on my Google Drive. My prompt was a simple one, as follows: 

@Google Drive  Find the papers that talk about the risks associated with AI 

Gemini quickly responded with the following. Finding this information by navigating through my folders and papers on my Google drive would have taken considerable time and by providing me a list of links to the individual documents on my Google drive, saved me even more time. 

Here are some of the papers I found in your Drive that discuss the risks associated with AI:

• X-Risks Paper.pdf discusses the potential for AI to pose existential risks to humanity, and suggests ways to mitigate these risks.

• practices-for-governing-agentic-ai-systems.pdf discusses the risks associated with agentic AI systems, which are AI systems that can pursue complex goals with limited direct supervision. The paper suggests a set of best practices for keeping these systems safe and accountable.

• gpt-4-system-card.pdf discusses the safety challenges presented by GPT-4, a large language model chatbot developed by OpenAI. The paper outlines the safety measures that OpenAI has taken to mitigate these risks.

• gpt-4.pdf also discusses the safety challenges of GPT-4, as well as the model's capabilities and limitations.

Items considered for this response. 

< list of links to the actual documents provided, but removed to keep this article brief>

Moreover, I can take this example further and chat with an individual document like this. 

@Google Drive what does the X-RisksPaper have to say about Scope Levels ?

Gemini’s response was comprehensive and correct  as follows: 

The document "X-RisksPaper.pdf" discusses Scope Levels in the context of AI risk analysis. The paper identifies three distinct scopes for considering AI risks:

1. AI System Risks: These risks concern the ability of an individual AI system to operate safely. Examples of AI system risks include anomalies, adversaries, and emergent functionality.

2. Operational Risks: These risks concern the ability of an organization to safely operate an AI system during deployment. Examples of operational risks include alarm fatigue, model theft, competitive pressures that undervalue safety, and lack of safety culture.

3. Institutional and Societal Risks: These risks concern the ability of global society or institutions that decisively affect AI systems to operate in an efficient, informed, and prudent way. Examples of institutional and societal risks include an AI arms race, incentives for creating AI weapons or using AI to create weapons.

The paper argues that all three scopes of risk are important to consider when evaluating the safety of AI systems. By taking a comprehensive approach to risk analysis, we can better understand the potential hazards posed by AI and develop more effective strategies for mitigating them.

I hope this summary is helpful!

If this was ChatGPT I would first have to locate my pdf in my file folders or through web search, then upload it to ChatGPT to search through it. So, if you want a second brain, with all that entails, just make Google drive your repository.

Seriously, I find this approach to storing and managing knowledge compelling, especially when you can combine this repository with the what’s contained  in my emails, which already resides on gmail, and You Tube.

I won’t show examples of @Gmail or @You Tube, because by now you get the idea and I leave it to explore these capabilities on your own. Although I will mention @Google Flights was an interesting example because I live in a small town east of Toronto, Canada. 

Before Gemini  when you searched for flight information, you have to specify Toronto (YYZ) as the starting point of any trip.  With Gemini, all I had to do is say '@Google Flights show me all the flights leaving today to Huatulco Mx' and Gemini determined that the nearest international airport was YYZ. 

Smarter Responses

The more time I spend with Gemini, the more I appreciated many of its responses. One of the differences you notice after using Gemini for a while is its propensity to suggest that it could provide a better answer with more information. I encountered this time and time again. In contrast, ChatGPT’s approach is to provide the best answer it can with the information you provided in your prompt. I prefer Gemini’s approach as it’s more conducive to learning. Consider the following prompt to Gemini and its responses.

Prompt: Determine the average city population using the top 5 largest Canadian cities 

Unfortunately, your request needs clarification on which definition of "city" you'd like to use for calculating the average population. Here are two common approaches:

1. Average city population within their municipal boundaries:

• Toronto: 2,794,356

• Montreal: 1,762,949

• Calgary: 1,306,784

• Ottawa: 1,017,449

• Edmonton: 1,010,899

Adding these populations and dividing by 5 gives an average of 1,378,467.4 inhabitants per city.

2. Average population of the larger census metropolitan areas (CMAs):

• Toronto CMA: 5,647,656

• Montreal CMA: 3,675,219

• Vancouver CMA: 2,426,160

• Calgary CMA: 1,305,550

• Edmonton CMA: 1,430,407

Adding these populations and dividing by 5 gives an average of 2,894,998.4 inhabitants per CMA.

Please specify which definition of "city" you prefer, and I'll be happy to provide the corresponding average population!

Gemini is suggesting that it could provide a more definitive answer if I chose one definition of a Canadian city over another. I told it to use the CMA definition.

Lets use CMA

Gemini’s response to my prompt is shown below:

Sure! Here is the average population of the top 5 Canadian CMAs:

City CMA Population

• Toronto 5,647,656

• Montreal 3,675,219

• Vancouver 2,426,160

• Calgary 1,305,550

• Edmonton 1,430,407

Export to Sheets

Noteworthy, is Gemini giving me the option to export the results to Sheets, highlighting its integration with Google Docs. Sheets is part of the free, web-based Google Docs Editors suite. 

Gemini's tendency to ask clarification or for more information on its own if Gemini thinks it can give me a better answer. I did not know there were two definitions for Canadian cities when I asked this question, but I do now.

I said earlier, I prefer Gemini’s approach because when you are exploring a topic that you don’t know much about, you don’t always know the best prompt to get the answers you are looking for. 

Summary

As I  said at the beginning  of this article, I am not ready to upgrade Google Gemini to a paid version, at least not yet. What I am ready to do is to make use of the free version of Google Gemini  as part of my AI toolkit especially where I need to leverage its integration with Google platform services like Google Docs and Google Drive. This means I will make more use of Google platform services, perhaps that is what Google is after.  

Gemini might not be a ChatGPT killer at this time, but it is a worthy competitor to Microsoft 365 with a Copilot.

I asked Gemini about Google Docs. Here is what Gemini told me. 

As of February 11, 2024, there are over 1 billion active users per month for Google Docs. This information comes from several sources:

• Exploding Topics: Their "Google Workspace User Stats (2024)" article claims over 1 billion active monthly users for Docs.

• Patronum: According to their "Key Google Workspace Statistics for 2023" report, Google Workspace itself (which includes Docs) has over 3 billion active users per month. While this figure encompasses individual and business users, it suggests a substantial Docs user base.

This could be the making of a good strategy for Google, growing its cloud services business, but whatever comes of it. Gemini should not be dismissed for what it has a lot to offer.