ChatGPT 4o vs Expert Analyst | Data Visualization: Who Does It Better?

Chris Dutton, Power BI expert and founder of Maven Analytics, goes up against OpenAI’s ChatGPT 4o to tackle a data visualization project.

The task?

Analyze a data set containing 120 years of Olympic history, and use data visualization to explore how the number of athletes from each country has trended over time.

Can ChatGPT alone solve this task or does it require the skills of an expert analyst?

⏱️ Timestamps ⏱️
0:00 Intro
0:27 Expert Analyst Chris Dutton
0:35 Data Visualization Task
0:55 ChatGPT Demo: Analysis
2:30 ChatGPT Demo: Visualization
4:37 Power BI Expert Demo: Analysis
10:27 Power BI Expert Demo: Visualization
13:10 Power BI Expert Demo: Final Visualization
14:16 Expert Analyst Thoughts
16:22 Recap
17:17 Who Does It Better?
17:25 Outro

🗣️ Follow us for more ⬇️
👉 Follow Chris:

👉Follow Alice:

👉 Follow Maven Analytics:

At Maven Analytics, we empower everyday people with life-changing data & AI skills.

[jasonrusso]

I'm a believer that a skilled analyst who is comfortable using AI is the best combination for speed and effective analysis. Great head to head challenge!

[investmentanalyst]

Rookie mistake, as an expert PowerBI user and Python user, I would have simply asked GPT to produce the code in python to perform exactly what need to do. Those with domain knowledge are the ultimate users of GPT. Because we know our process in detail, we have provide GPT with a high quality prompt to produce exactly what I'm looking for. I could have produced a better chart that explained the data a bit better in Python. The only reason I'd graph something in PowerBI is if Management wants to "play" with the data. If, I'm producing for internal use of other domain knowledge users, python is more powerful.

[jbpauler]

This is awesome. ChatGPT is decent, but still has a long way to go. I'll take Chris Dutton over the machines all day!!

[andreag]

Chris Dutton did it better: he considered multiple events per athlete, the duality between summer and winter games, and historical events that occurred in the twentieth century. Alice Zhao could have guided ChatGPT more using the prompt, but the AI ​​alone, although extremely fast, is not able to grasp the nuances

[ramabg]

I would love to see how expert data analyst leverage GenAI to make data analyst process faster and better. Like asking ChatGPT why in certain year Olyompic participan is lower or asking chat GPT to make a list of countries based on regional, continent, developing/ develop.

That way, some tedious job can be offloaded to GenAI hence the expert can focus more on creative work.

[rselvarajanMBA]

What no one realised is that when you click on the star like icon, Chat GPT shows the actual python code it used to analyse and create the visualisation. It is quite useful and you can modify it to get far more detail results.

[osPA]

This is a great video on how ChatGPT, Copilot, etc. can tee up the data for an analyst to bring it home! Really enjoyed watching this!!!

[cathyfrankland]

Fascinating, thank you so much for this interesting example of AI vs. Human. Also, well done, Chris, impressive (as always!).

[ateekagarwal]

Well said… Data analyst/ human can interpret data more granulated manner by analysing what’s the cause of it
While AI is used as a supplements…
Highly impressed by AI but at the end we have to actually finding it’s actual RC(root cause) and it’s recommendations according to dynamic market trends.

[AbdifatahGobe]

I enjoyed watching from the beginning. Alice could have done more with the prompts, but GenAI appears to require human guiding. I also considered premium AI versions such as GPT 4o and the use cases for which they were created, and whether they could have provided a much more clear and brief analysis.

Chris Dutton is at the top of the perk list for data analysis. It would have been appropriate to compare the AI to a typical data analyst. Special thanks to Chris for the insights! We both, me and the AI, have a lot to learn from you.

[BelgranoCAB]

I had love this video. Greetings from Argentina!

[saifiiui]

The slicing of the season dimension was really an important insight in data that chatgpt ignored 😊.
The main difference between AI and the human Evolutionary aspect is the way humans have feelings. So chatgpt can never replace the human mindset but surely can assist the human generations to come.

[Maouww]

Really cool experiment - the conclusions are spot on.
But if you want a truly interactive data plot, you can ask for python and customise from there.

Still, at the end of the day, chatgpt excels at repetitive rudimentary tasks (coz that's what it's trained on) but our expertise is always needed to:
1. Direct it
2. Tailor specific needs

Interestingly, There's a paper published recently showing that AI outperforms humans in QA which is honestly one less tedious thing for us to think about.

[malcolmvanhilten]

I think this excercise was a great experiment. My conclusion is that Ai will most likely take over the role of building dash boards and the tech stuff. It got to the point of the first viz, as the (very senior) analyst, 80% quicker. That meant that the skill of building PowerBi and data tables etc is fast becomming obsolete, which is good news. The analyst was able to demostrate more value (much more) in recognising the distinct count and the change in the games frequency. So in the future if your value add is limited to building dash boards and you dont really understand the data, then your job is very much under threat. Lets be honest if you see data as just dots on a page (exactly like ChatGPT did in this example) then you have little value to offer going forward.

[skd]

ChatGPT can make 2+2 mistakes so the human always does it better. ChatGPT can say sorry in many ways though.

[kwobafredrick]

Impressed on how Chris went about it 😊

[faustopf-.]

Wow! Insane skills and a super interesting topic! Love it

[veselinkobg]

Much respect for the expert analyst, looking to become one someday, but if the initial promt was "How many DISTINCT athletes are from each country, each year?", ChatGPT gives the correct answer

[ezscootrr]

can the gpt connect directly to servers?

[rupertllavore]

Interesting Concept! Maybe you should increase the odds By using atleast Some good Prompt engineering skills for Chatgpt