Constructing an Excel dashboard normally means spending a complete afternoon on format design. So, I pitted ChatGPT towards Gemini to see which instrument automates the grunt work finest—and just one got here near what I hoped for.
For each fashions, I used their free tiers—ChatGPT Free and Gemini Flash—so the outcomes replicate what most individuals would truly expertise with out paid options or extensions. I additionally made positive I gave every instrument the identical dataset, immediate, and constraints.
Here is the way it went.
I requested each fashions to generate a accomplished Excel file from scratch
Direct workbook era broke down in apply
I began by asking each instruments to construct a whole Excel dashboard from scratch, utilizing a modified model of Microsoft’s Financial Sample dataset, to see whether or not they might produce a working file I might open and use instantly.
The immediate started by telling each fashions to behave as Excel dashboard designers, and I requested them to make use of a strict three-sheet architecture (Knowledge, Processing, Dashboard), relying solely on native Excel formulation and features. I additionally instructed them to create KPI playing cards for Whole Gross sales, Whole Revenue, and Whole Items Bought, produce PivotTables, insert three charts, embrace a slicer linked to the complete dashboard, and apply an expert design utilizing navy accents and muted greys.
Sadly, the direct file era strategy fell flat.
Gemini produced a usable workbook that accurately rebuilt the dataset—albeit with none formulation. It generated KPI playing cards and charts that mirrored precise values, though the colours did not meet my transient, and it relied on hard-coded outputs within the processing layer as a substitute of a totally reusable PivotTable system.
ChatGPT additionally produced a workbook, but it surely arrived in a far much less full state. The dataset was not totally reconstructed, the processing sheet contained ineffective placeholders, and the dashboard playing cards displayed zeroes and clean charts resulting from lacking information connections.
Whereas each outputs—significantly the one generated by Gemini—regarded visually promising at a look, neither delivered a dependable, reusable dashboard construction. I shortly realized that asking these instruments to instantly generate the advanced spreadsheet file, even with comprehensive prompts, was the fallacious strategy for this type of process.
Quiz
Synthetic intelligence fundamentals
Trivia problem
From chatbots to neural networks — learn how a lot you actually find out about AI.
IdeasHistorical pastInstrumentsEthicsFashions
What does the time period ‘machine studying’ most precisely describe?
Right! Machine studying is a department of AI the place programs enhance routinely via expertise and publicity to information. As a substitute of being explicitly programmed for each process, these programs establish patterns and make selections with minimal human intervention.
Not fairly. Machine studying refers to programs that be taught from information to enhance their efficiency over time. It is much less about bodily motion or precise mimicry and extra about discovering patterns in giant datasets to make predictions or selections.
Who’s extensively credited with coining the time period ‘synthetic intelligence’ in 1956?
Right! John McCarthy coined the time period ‘synthetic intelligence’ on the well-known Dartmouth Convention in 1956, which is taken into account the founding occasion of AI as a proper discipline of analysis. He later invented the Lisp programming language, which turned a staple in early AI improvement.
Not fairly. Whereas Alan Turing, Marvin Minsky, and Claude Shannon have been all AI pioneers, it was John McCarthy who coined the time period ‘synthetic intelligence’ on the Dartmouth Convention in 1956. McCarthy went on to form the sector enormously all through his profession.
What kind of AI mannequin powers in style chatbots like ChatGPT?
Right! ChatGPT and related chatbots are powered by giant language fashions, or LLMs. These fashions are educated on monumental quantities of textual content information and be taught to foretell and generate human-like language, making them able to dialog, writing, and reasoning duties.
Not fairly. ChatGPT is constructed on a big language mannequin (LLM). Whereas resolution timber and Bayesian classifiers are actual AI instruments, they’re used for a lot less complicated duties. CNNs are nice for picture recognition however aren’t designed for open-ended language era.
What’s ‘overfitting’ in machine studying?
Right! Overfitting occurs when a mannequin learns the coaching information too effectively — together with its noise and quirks — after which fails to generalize to new, unseen information. It is like a pupil who memorizes apply examination solutions however cannot deal with totally different questions on the true take a look at.
Not fairly. Overfitting describes a mannequin that has realized the coaching information so particularly that it performs poorly on new information. It is one of the widespread challenges in machine studying and is addressed via methods like cross-validation and regularization.
What’s ‘AI bias’ mostly referring to?
Right! AI bias refers to systematic errors or unfair outcomes that come up when a mannequin is educated on skewed, incomplete, or unrepresentative information. For instance, facial recognition programs have been proven to carry out worse on darker pores and skin tones resulting from biased coaching datasets, elevating severe moral considerations.
Not fairly. AI bias is about systematic, typically dangerous unfairness baked right into a mannequin’s outputs, normally resulting from skewed coaching information or flawed design decisions. It is a main moral concern in areas like hiring algorithms, prison justice instruments, and medical diagnostics.
What does ‘GPT’ stand for in AI mannequin names like GPT-4?
Right! GPT stands for Generative Pre-trained Transformer. ‘Generative’ means it may possibly create new content material, ‘pre-trained’ means it was educated on a big dataset earlier than being fine-tuned, and ‘Transformer’ refers back to the neural community structure that made trendy LLMs attainable.
Not fairly. GPT stands for Generative Pre-trained Transformer. The Transformer structure, launched in a landmark 2017 paper referred to as ‘Consideration Is All You Want,’ revolutionized pure language processing and laid the groundwork for as we speak’s highly effective AI chatbots.
Which of the next finest describes ‘deep studying’?
Right! Deep studying is a subset of machine studying that makes use of synthetic neural networks with many layers — therefore ‘deep’ — to mannequin advanced patterns in information. It is the know-how behind picture recognition, voice assistants, and most trendy AI breakthroughs.
Not fairly. Deep studying makes use of multi-layered neural networks impressed loosely by the human mind. The ‘depth’ refers back to the variety of layers within the community, and extra layers typically enable the mannequin to be taught extra advanced and summary representations of information.
What was the identify of the IBM AI system that famously defeated chess champion Garry Kasparov in 1997?
Right! IBM’s Deep Blue defeated world chess champion Garry Kasparov in a six-game match in 1997, marking a landmark second in AI historical past. It was the primary time a pc beat a reigning world chess champion underneath normal match circumstances, stunning the world.
Not fairly. The IBM system was referred to as Deep Blue. Watson is IBM’s later AI recognized for profitable Jeopardy!, whereas AlphaGo is Google DeepMind’s system that mastered the board recreation Go in 2016. HAL 9000, after all, is the fictional AI from Stanley Kubrick’s 2001: A House Odyssey.
Your Rating
/ 8
Thanks for enjoying!
So, as a substitute of asking for a completed workbook, I modified my technique in order that each fashions would produce a full automation blueprint that I might comply with to construct the Excel dashboard manually. I used the very same dataset and constraints as earlier than however tweaked the immediate to request particular directions, cell coordinates, shade hex codes, and aspect sizing.
This shift modified the character of the comparability by exposing how exactly every mannequin might translate summary necessities into concrete implementation steps.
- OS
-
Home windows, macOS, iPhone, iPad, Android
- Free trial
-
1 month
Microsoft 365 consists of entry to Workplace apps like Phrase, Excel, and PowerPoint on as much as 5 gadgets, 1 TB of OneDrive storage, and extra.
Gemini’s blueprint was exact and straightforward to implement
Clear coordinate mapping made guide meeting a breeze
Gemini’s blueprint translated cleanly into Excel as a result of it delivered precise cell references and particular chart ranges. I significantly appreciated that it supplied direct supply references for charts, reminiscent of particular sheet ranges, which means I might copy them straight into Excel’s chart supply fields quite than manually choosing PivotTable outputs.
It dealt with KPI formulation cleanly, ensured PivotTable outputs aligned with the dashboard format, and persistently mapped coordinates. The complete construction felt steady as soon as I had completed the guide construct.
Gemini’s visible system additionally labored. It used the requested palette with navy accents, correctly structured the KPI playing cards, and urged a horizontal slicer layout that considerably improved spacing.
Nonetheless, the blueprint nonetheless wanted minor fixes. The slicer dimensions have been too small to completely show its textual content buttons, the doughnut chart left one section with default neon coloring as a result of the mannequin solely informed me to restyle 5 of the six classes, and chart titles have been lacking. I additionally wanted to make some minor changes to chart parts and vertical spacing to enhance readability.
That stated, none of those points affected core performance, and it solely took about two minutes of fast tweaks to show the outcome into one thing I used to be proud of.
ChatGPT’s blueprint required in depth guide interpretation
Obscure directions and damaged formulation added friction
ChatGPT coated the identical core parts: formulation, PivotTable setup, chart placement, and slicer configuration. Nonetheless, it gave me far much less exact coordinate mapping, resulting in some confusion when translating directions into Excel.
For instance, it described format areas utilizing normal phrases like “Background: White; Unused space: Smooth Grey Fill” with out defining clear boundaries between them, and following the precise placement directions for the slicer meant that it was obscured by one of many charts as soon as I had completed assembling the whole lot.
The charts additionally felt vertically compressed, which diminished readability and demanded extra guide changes than Gemini’s extra spaced format.
AI hallucination crept into ChatGPT’s output as effectively. It included an Govt Perception part designed to generate automated textual content commentary—however this characteristic failed in apply as a result of the underlying formulation returned errors or empty outcomes, which means I needed to both rebuild or take away the extra part solely.
A number of formatting points required correction, together with default doughnut chart colours that weren’t overridden, a poor KPI card hierarchy during which labels and values lacked distinct visible weight, and tight spacing that induced parts to crowd each other. ChatGPT additionally solely informed me to take away the background in a single chart, quite than all three.
On the plus aspect, it explicitly prompted me to rename my chart objects—a key element Gemini ignored. And its PivotTable directions have been typically simpler to comply with than Gemini’s, though each fashions nonetheless assumed some present Excel information.
Whereas none of ChatGPT’s points have been catastrophic, the sheer variety of them meant that it took for much longer to repair the outcome than it did for the Gemini output.
The true hole seems when mapping parts to the grid
ChatGPT and Gemini each accurately structured the workbook into Knowledge, Processing, and Dashboard sheets, supplied legitimate KPI formulation for Gross sales, Revenue, and Items Bought, and clearly defined PivotTable creation and slicer connectivity.
The essential variations turned apparent as I applied the steps. Gemini supplied reusable, instantly copyable parts like express cell ranges and chart sources, however ChatGPT relied on generic directions and struggled with formatting. In consequence, Gemini’s assured output required fewer corrections and locked into place cleanly, whereas ChatGPT’s much less exact blueprint meant I spent for much longer refining spacing and visible construction earlier than I might truly use the workbook.
And the winner is…
Neither mannequin delivered a totally automated Excel dashboard from the preliminary workbook request. As soon as I switched to the blueprint strategy, nonetheless, the variations turned clearer. Gemini produced the extra exact and usable outcome and required solely minor tweaks, whereas ChatGPT’s skewed layouts and hallucinations meant its output wanted extra guide refinement earlier than reaching the identical stage of polish.
In sensible phrases, the experiment confirmed that the important thing distinction was not uncooked functionality however instruction precision. As a result of Gemini supplied clearer formulation, precise grid coordinates, and extra exact format directions, it was the one instrument that really diminished the quantity of guide effort and time required to construct the Excel dashboard I needed.
I tested ChatGPT, Claude, and Gemini for months, and the winner surprised me
The AI race is simply too quick to maintain up with—and that is okay
Source link
