Test-driving Google’s Gemini-Exp-1206 model in data analysis, visualizations

one Google’s latest experimental model Gemini-Exp-1206, Demonstrated potential to alleviate one of the most intractable aspects of any problem analyst’s Job: Keep their data and visualizations in perfect sync and provide compelling narratives, No need to work all night.

Investment analysts, junior bankers, and advisory team members aspiring to partner positions know when approaching their roles that: long timeweekends, and the occasional all-nighter may give them an internal advantage when it comes to promotions.

They spend a lot of time completing high-level data analysis and creating visualizations to enhance Compelling storyline. To make matters even more challenging, each bank, fintech and advisory firm, including JPMorgan Chase, McKinsey and PwC, has unique formats and conventions for data analysis and visualization.

VentureBeat interviewed members of the internal project teams whose employers hired the companies and assigned them to the project. Employees who work on consultant-led teams say creating visuals that can compress and integrate large amounts of data is an ongoing challenge. Teams of consultants often work through the night, going through at least three or four iterations of visuals for briefings before one is finalized and ready for a board-level update, one person said.

Compelling use cases for test driving Google’s latest models

Used by process analysts to create presentations that support storylines with solid visualizations and graphics, there were so many manual steps and duplication that it proved a compelling use case for testing Google’s latest model.

When launching the model in early December, Google’s Patrick Kane Wrote”, “Whether you are tackling a complex coding challenge, solving a math problem for school or a personal project, or providing detailed multi-step instructions to develop a customized business plan, Gemini-Exp-1206 will help you complete complex tasks more easily. Task. Google noted that the model’s performance improved on more complex tasks, including mathematical reasoning, coding, and following a sequence of instructions.

VentureBeat conducted a thorough test of Google’s Exp-1206 model this week. We created and tested over 50 Python scripts in an attempt to automate and integrate analysis with intuitive, easy-to-understand visualizations that simplify the complex data being analyzed. Given the dominance of hyperscalers in today’s news cycle, our specific goal is to analyze a given technology market while creating supporting tables and advanced graphics.

Through more than 50 different iterations of the proven Python script, our findings include:

• The greater the complexity of the Python code request, the more the model has to “think” and try to predict the desired outcome. Exp-1206 attempts to predict what a given complex prompt will require, and will alter the content it produces by the slightest changes in the prompt. We see this in how the model alternates between table-type formats placed directly above the very large market analysis spider chart we created for testing.

• Force models attempt complex data analysis and visualization and produce Excel files, providing multi-tab spreadsheets. Without being asked for an Excel spreadsheet with multiple tabs, Exp-1206 created one. The requested primary tabular analysis is on one tab, the visualization is on another tab, and the secondary tabulation is on a third tab.

• Let the model iterate over the material and recommend the 10 visualizations it thinks are the best fit for the material, providing helpful and insightful results. To reduce the time required to create three to four iterations of slides before board review, we force the model to generate multiple concept iterations of the image. These can be easily cleaned up and integrated into presentations, saving a lot of manual work in building charts on slides.

Push Exp-1206 to complex, layered tasks

VentureBeat’s goal is to see how far the model can go in terms of complexity and layered tasks. Its performance in creating, running, editing, and fine-tuning 50 different Python scripts shows how quickly the model attempts to capture nuances in the code and react immediately. Models adjust and adjust based on real-time history.

Results of running Python code created using Exp-1206 Google Shown is the subtle granularity that extends to shading and translucency across layers in an eight-point spider diagram designed to show a comparison of six hyperscale competitors. We require that the eight properties that Exp-1206 identifies and anchors spider graphs remain consistent across all hyperscalers, but that the graphical representations vary.

Battle of the Hyperscale Enterprises

We selected the following hyperscale services for testing and comparison: Alibaba Cloud, Amazon Web Services (AWS), Digital Realty, Equinix, Google Cloud Platform (GCP), Huawei, IBM Cloud, Meta Platforms (Facebook), Microsoft Azure, NTT Global Data Center , Oracle Cloud and Tencent Cloud.

Next, we wrote 11-step prompts with over 450 words. The goal is to see how well Exp-1206 handles sequential logic without losing its place in complex multi-step processes. (You can read the tips in the appendix at the end of this article.)

Next we submitted the tip Google Artificial Intelligence Studioselect the Gemini Experimental 1206 model, as shown in the figure below.

Next, we copy the code into Google Colab and save it into a Jupyter notebook (Hyperscaler Comparison – Gemini Experimental 1206.ipynb), and then execute the Python script. The script works perfectly and creates three archives (indicated by the red arrows in the upper left corner).

Hyperscaler comparison analysis and charts – less than a minute

The first series of instructions in the prompt asks Exp-1206 to create a Python script that will compare 12 different hyperscale processors by product name, unique features and differentiating factors, and data center location. Below are the results for the Excel file requested in the script. It took us less than a minute to format the spreadsheet to shrink it to fit in the columns.

The next series of commands asks for a table with the top six hyperscalers compared at the top of the page and on the spider graph below. Exp-1206 chose to represent the data in HTML format and created the page below.

The final sequence of prompt commands focuses on creating a spider graph comparing the first six hyperscales. We tasked Exp-1206 with selecting eight comparison standards and completing the drawing. This series of commands is translated into Python, and the model creates the archive and serves it in a Google Colab session.

Models built to save analysts time

VentureBeat learned that on a daily basis, analysts are constantly creating, sharing, and fine-tuning libraries of hints for specific AI models with the goal of streamlining reporting, analysis, and visualization for the entire team.

Teams working on large consulting projects need to consider how a model like the Gemini-Exp-1206 can greatly improve productivity and alleviate the need for 60+ hour weeks and occasional all-nighters. A series of automated prompts can do the exploratory work of looking at relationships in the profile, allowing analysts to produce visuals with greater certainty without spending undue time.

appendix:

Google Gemini Experiment 1206 Prompt Test

Write a Python script to analyze the following hyperscalers that have declared global infrastructure and data center presence for their platforms, and build a comparison table that captures the significant differences in each approach to global infrastructure and data center presence.

The first column of the table is the company name, the second column is the name of each of the company’s hyperscalers with global infrastructure and data centers, and the third column is what makes its hyperscalers unique and drills down into the most differentiating ized feature functions, the fourth column is the location of each hyperscale data center at the city, state, and country levels. Include all 12 very large programs in an Excel file. Do not scrape the web. Generates an Excel file of the results and formats the text in the Excel file to remove any brackets ({}), quotation marks (‘), double asterisks (**), and any HTML code to improve readability. Name the Excel file Gemini_Experimental_1206_test.xlsx.

Next, create a table that is three columns wide and seven columns deep. The first column is titled “Hyperscale,” the second column is titled “Unique Features and Differentiators,” and the third column is titled “Infrastructure and Data Center Locations.” Make the column headers bold and center them. The names of hyperscalers are also bolded. Double check to make sure that the text within each cell of the table wraps and does not cross into the next cell. Adjust the height of each row to ensure that all text fits into its intended cell. This table compares Amazon Web Services (AWS), Google Cloud Platform (GCP), IBM Cloud, Meta Platforms (Facebook), Microsoft Azure, and Oracle Cloud. Center the table at the top of the output page.

Next, we use Amazon Web Services (AWS), Google Cloud Platform (GCP), IBM Cloud, Meta Platforms (Facebook), Microsoft Azure, and Oracle Cloud as examples to define the eight most differentiated aspects of this group. Use these eight different facets to create a spider diagram comparing the six hyperscales. Create a large spider diagram that clearly shows the differences between the six hyperscalers, using different colors to improve their readability and the ability to see the outlines or footprints of the different hyperscalers. Be sure to title your analysis “What Most Differentiates Hyperscalers,” December 2024.

Add a spider graphic to the bottom of the page. Center the spider graphic below the table on the output page.

The following are hyperscale services included in Python scripts: Alibaba Cloud, Amazon Web Services (AWS), Digital Realty, Equinix, Google Cloud Platform (GCP), Huawei, IBM Cloud, Meta Platforms (Facebook), Microsoft Azure, NTT Global Data Center, Oracle Cloud, Tencent Cloud.