Digital

When I Tell ChatGPT It Needs a History Lesson, It Totally Convinced Me

uTech News2024-12-1807 mins

AI’s Limits Reverse an Academic Map Quest

After a week‑long experiment that highlighted the gaps of both ChatGPT and Google’s Gemini, I realised that enthusiasm for AI must be tempered by realism.

Why the Test Matters

Both platforms produce impressive text but stumble on visual tasks.
One project that seemed trivial became a pain point.
My daughter’s dissertation hovered over a boundary comparison that should be straightforward.

The Project Challenge

My daughter, nearing the end of her doctorate, asked AI to generate a map overlaying the borders of the Byzantine Empire at two pivotal moments:

379 AD – The start of Theodosius the Great’s reign.
457 AD – The final year of Marcian’s rule.

She needed a single map that clearly displayed both borders, styled like a historical map, and highlighted major cities.

What AI Delivered (and Didn’t)

Using the “Deep Research” option, ChatGPT produced a thorough text analysis that explored border changes, city importance, and related events.

When I asked AI to convert that analysis into a visual map it fell short. The platform can articulate past events but fails to create an accurate, easy‑to‑read map in a single pass.

Takeaway

While AI tools like ChatGPT and Gemini excel at generating detailed narratives, they are not yet capable of handling complex visual tasks—at least not without significant manual intervention.

This experience reminds us that grand expectations must be matched with the reality that AI is still in progress, especially when it comes to producing high‑quality, data‑rich visuals.

ChatGPT’s Geographic Folly

It’s easy to find a short story that misplaces Rome on the Italian peninsula and drifts Antioch into Europe. What makes it truly maddening is the fact that the names and locations appear correct in the source text.

Stubborn Attempts at Fixing the Errors

After a series of patient requests for better spelling and accurate placement of well‑known cities, I was forced to tell the model that its output was garbage and to throw up my hands. ChatGPT replied thank you for your candor and you are right to expect better, but the improved quality was still unacceptable.

Google Gemini’s Even Worse Results

Rome appears in the middle of the Iberian Peninsula.
Antioch shows up three or four times across Europe.
Many other place names are complete fabrications.

When the same query was issued to Google Gemini, the output was even more disastrous. The resulting image‑based map shows Rome in the wrong country and cherry‑picks Antioch across an unrealistic European landscape. The names and locations are completely out of place and look like they were lifted from a fantasy novel.

Conclusion

Both models are poor at matching historical geography and will generate fictitious city names when asked for the precise placement of ancient and medieval locations. Until the models can be fine‑tuned with accurate, historical datasets, users will continue to see ill‑structured maps and misplaced cities.

AI Missteps Highlight the Need for Human Oversight

During a casual chat, I shared a frustrating incident of image manipulation. A friend recounted a similar mishap: they asked ChatGPT to overlay “Mahalo from Hawaii 2025” on a snapshot from a modest offsite gathering. Rather than a simple text addition, the AI altered the entire scene—slimming individuals, swapping men for women, and even reassigning an Asian face to a Caucasian appearance.

Another anecdote involved an AI‑crafted biography that mentioned twin children—details the subject does not possess—and even cited a nonexistent source. The consequences were unsettling.

Lessons Learned

Trust but Verify—a principle long championed by Ronald Reagan.
AI as Tools, Not Guardians—we must handle their output, assess its authenticity, and decide whether to adopt it.
Performance is Uneven—the engines excel in some domains but flounder in others, such as image mapping.
Human Judgment Remains Essential—we cannot hand over the keys and walk away.

Looking Ahead

While the machines may eventually assume greater responsibilities, the present day demands a cautious approach. AI can enhance our workflows, but the final verdict must rest in human hands.