Is This What Superintelligence Feels Like? – Inside o3’s Geoguessing Feat

AI guessed a photo location with near-perfect accuracy — down to the street.

Not long ago, a photo shared on social media caused a stir in the AI world. In the image, there wasn’t anything famous. It was just a regular street view. But what happened next amazed people.

An AI model from OpenAI, called o3, guessed the photo’s location with stunning accuracy. It didn’t just name the country; it pinned the exact street.

This AI wasn’t a regular chatbot or search engine. o3 is OpenAI’s most advanced reasoning model so far. Unlike earlier models that gave fast answers, o3 is built to think more deeply.

While o3’s performance on tests was impressive, what really caught people’s attention was its skill at geoguessing. It’s a game where you figure out where a photo was taken, just by looking at it. Some people posted photos, even personal or edited ones, and o3 still figured out the location.

That’s when the internet exploded with debate. Some users wondered: Is this what superintelligence feels like?

This article looks at that question. Is o3’s skill a hint of a future where AI thinks better than humans? Or is it just really good at one narrow task?

What Is o3 and What Can It Do?

*All models are evaluated at high ‘reasoning effort’ settings—similar to variants like ‘o4-mini-high’ in ChatGPT.*

OpenAI’s o3 is a powerful new AI model built for deep thinking.

Unlike older models that focus on speed, o3 is designed to slow down and reason carefully. OpenAI calls it a “reasoning model”, meaning it can break big problems into smaller steps and solve them logically.

One of its biggest strengths is how well it understands images.

o3 can study photos, diagrams, even rough sketches, and pull out meaningful details — like plant types, road signs, or the angle of a shadow. The release note from OpenAI reads:

For the first time, these models can integrate images directly into their chain of thought. They don’t just see an image—they think with it. This unlocks a new class of problem-solving that blends visual and textual reasoning, reflected in their state-of-the-art performance across multimodal benchmarks.

o3 can also use external tools to solve problems.

It can write and run code, zoom into parts of an image, or search the web for more clues. It decides which tools to use and when, combining results to reach better answers. This kind of “tool-based reasoning” is a big leap forward.

It’s already scored high on tests in areas like coding, logic puzzles, and advanced reading. That broad skillset helps it perform well on more specific tasks — including geoguessing.

The Geoguessing Process

Geoguessing means figuring out where a photo was taken just by looking at it. o3 is surprisingly good at this, but only when guided by a detailed prompt; often written by a human.

The prompt tells o3 to start with pure observation: colors, shapes, signs, shadows; no guessing yet. Then it sorts clues into categories: what’s the climate, what side of the road do cars drive on, what kind of buildings are there?

Next, o3 lists a few possible locations and explains its thinking. It might use tools to zoom in on a license plate or run a web search. It tests each guess, looking for flaws or missing details. Before choosing a final answer, it actively tries to disprove itself — just like a careful human would.

In one case, it looked at a beach town photo and, after analyzing flowers, fences, and hills, zoomed in on a blurry license plate. It correctly guessed Cambria, California — and even listed a backup guess.

This smart process shows how o3 uses visual clues, memory, logic, and tools together. But it also shows that it works best with a clear plan. On its own, it may not perform this well. That’s a key point here.

o3 vs. The World in GeoGuessr

To know how good o3 really is, we have to ask: Can it beat the best human players? One match gave us a clear answer — well, kind of.

The Sam Patterson Match

In a five-round head-to-head game, o3 took on Sam Patterson, a top-ranked GeoGuessr player. Patterson is no amateur; he plays in the Master I division. Yet, o3 won overall in this match, scoring 22,197 points to Patterson’s 21,379.

But there’s a twist: Patterson actually won more rounds — 3 out of 5.

o3 came out ahead only because it nailed two rounds with extreme precision, sometimes guessing within a few hundred meters of the actual spot. In one round in Norway, it scored just a bit more than Patterson. In Tunisia, the difference was just four points.

Table 1: o3 vs. Human GeoGuessr Master (Sam Patterson) – Comparative Performance

Metric	OpenAI o3	Sam Patterson (Master I)	Notes/Source
Overall Match Score	22,197	21,379	o3 won narrowly overall
Rounds Won (out of 5)	2	3	Patterson won more individual rounds
Accuracy (Winning Rds)	High precision (e.g., within hundreds of meters)	High, but less precise than o3’s best	Explains o3’s score advantage despite fewer round wins
Consistency	Variable; prone to errors/distractions	Likely more consistent	Based on critiques of o3’s errors and focus issues
Speed	Can be slow, methodical/repetitive	Generally faster, more intuitive	Based on observations of o3’s process and comparisons
Tool Use	Integrated web search, image analysis	Unassisted (standard) / Google sometimes	o3 leverages external digital tools seamlessly, human use of Google varies
EXIF Data Handling	Detected and ignored fake data	N/A (Human intuition)	Shows primacy of visual reasoning over potentially misleading metadata

Patterson’s takeaway? o3 plays at a very high level, possibly better than many top humans. But it’s not unbeatable. Other expert players, given the same images, sometimes outscored it.

Where o3 Has the Edge

o3 stands out in a few big ways:

Extreme precision: When it guesses right, it really nails it; thanks to detailed clues like shadow direction or exact plant types.
Massive memory: o3 likely draws from a huge training set that includes global photos and facts. It may recognize things even expert humans don’t.
Smart web searches: It can read a business name on a van, search it, and find the exact town — like it did with a taxi company in Austria.
Step-by-step analysis: When prompted, it checks everything — roads, signs, shadows — in a methodical way that sometimes beats human instincts.

Where Humans Still Win

But humans still do some things better:

Speed and focus: Humans can quickly spot the most important clue. o3 sometimes wastes time zooming in on ads or repeating tasks.
Consistency: o3 has highs and lows. A human expert is often more reliable round to round.
Common sense: Humans understand the real world better. AI can miss obvious hints or get confused by strange data.
Adapting on the fly: Humans can change strategies mid-game. o3 sticks to its prompt, which limits its flexibility.

Is o3 Superhuman at GeoGuessr?

Not quite. While o3 can outperform some humans in certain rounds, it doesn’t dominate.

Its win over Patterson came from a few pinpoint guesses, not overall mastery. It still struggles with simple scenes or odd images. And while it uses tools impressively, its reasoning isn’t perfect.

So, is o3 superhuman? No — but it’s definitely elite. It competes at the top level of GeoGuessr, showing off precision and power that are impressive, but not magical.

Is This Superintelligence? Let’s Define It

When o3 nailed the location of a personal photo with near-perfect accuracy, some people called it a glimpse of superintelligence. But is it really?

To answer that, we need to understand what superintelligence actually means and how far o3 still is from that mark.

https://twitter.com/Simeon_Cps/status/1918766794304614877

What Superintelligence Really Means

The term comes from philosopher Nick Bostrom, who defines superintelligence as any intellect that far exceeds human thinking in almost every area — not just one task like geoguessing, but across the full range of reasoning, planning, learning, and problem-solving.

Bostrom describes several types:

Speed superintelligence, which thinks like a human but much faster.
Collective superintelligence, made up of many minds working together.
Quality superintelligence, which can think in ways humans simply can’t.

True superintelligence wouldn’t just be faster or more accurate. It might reason in ways we don’t understand, or see patterns that are invisible to us. It would likely invent new knowledge, strategies, or tools; not just use the ones it’s given.

o3 Isn’t That — And Real Users Know It

o3, on the other hand, is a very good narrow AI. It excels at one task — geoguessing — when it has the right instructions. Its reasoning process is impressive but not mysterious. It looks at shadows, fences, plants, and signs. It cross-checks with web searches. It follows clear steps, just like a smart human might.

And it doesn’t do this on its own.

This gap between appearance and reality is something Reddit users picked up on quickly. After the viral demo, many commenters noted that o3’s success was largely due to an extremely detailed prompt; one that functioned more like a custom program than a simple question.

One user commented: “She had to use o3 many times to learn its drawbacks. The human, for now, is in place of a true AI metacognitive feedback loop.” Another added, “This isn’t magic. It’s a plugin written in prompt form, designed through trial and error.”

Others tested the model themselves and got mixed results. “I gave it a photo of my front yard,” one person wrote, “and it was off by 2,000 miles.” Another said, “It usually mentions the right region in its reasoning but still lands far away.” Many agreed: o3 can be accurate, but only with a lot of human scaffolding.

This matters because true superintelligence, as Bostrom defines it, should not depend on a human writing detailed instructions. It would create its own strategy and adapt on the fly. o3 doesn’t do that — it executes a plan, but doesn’t invent one. As one commenter put it, “Following a recipe isn’t the same as creating one.”

Not a Glimpse of the Future — Yet

So is o3 superintelligent? Not even close. It’s a brilliant tool doing a single job very well, under the guidance of a human expert. That’s not a knock — it’s a big achievement. But calling it superintelligence blurs the line between a tool that follows instructions and a mind that writes its own playbook.

The real takeaway here isn’t that o3 is smarter than us. It’s that when you combine a reasoning AI with smart human guidance, the results can feel extraordinary. But feeling isn’t the same as being — and understanding that difference is key.

Why o3 Feels Like Superintelligence Anyway

Even though o3 isn’t truly superintelligent, it’s easy to see why it feels that way; especially when it guesses a location with near-perfect accuracy. Watching it work can feel like watching magic.

1. It Explains Its Thinking

One reason o3 seems so smart is that it shows its reasoning. When guided by a good prompt, it walks through the process step by step:

It lists what it sees — buildings, signs, shadows.
It explains what those clues might mean.
It gives multiple options and says how confident it is.
It even tells you when it’s not sure.

This kind of clear, thoughtful explanation feels very human. But it’s really just following a set of smart instructions.

2. It Uses Tools Like a Detective

o3 also feels impressive because it uses tools in clever ways. It zooms in on blurry details, searches the web for business names, and compares patterns. For example, it might see a license plate, clean it up with code, and then match it to a U.S. state.

It’s not guessing blindly; it’s investigating. And that looks a lot like genius from the outside.

3. It Pays Attention to Tiny Clues

o3 notices things most people would miss — like the type of grass in a field or the angle of a tree’s shadow. It can even tell which side of the road cars drive on just by looking at signs and poles. This level of detail, pulled together fast, makes it feel superhuman.

But what’s really happening is a guided checklist. It’s not alien intelligence.

4. The “Shock” of Progress

Just a year ago, people thought this kind of AI reasoning wasn’t possible. Now it’s happening in public demos. That sudden leap forward creates what some call “AI shock” — the feeling that something big has changed, fast.

That feeling can be thrilling. But it can also make us overstate what the AI is really doing.

The Real Signal — Not Superintelligence, but a Super Tool

What o3 shows us isn’t the rise of some all-knowing mind. It’s something more grounded, but still powerful: a glimpse into the future of AI as a real-world problem solver.

Most AI systems today are built to respond to inputs. You ask, they answer. But o3 works differently. When it’s solving something like geoguessing, it doesn’t just guess and move on. It pauses, plans, checks its work, and uses tools to improve its answer. This process looks more like reasoning than reacting.

This kind of behavior is part of a growing shift in AI development. Instead of simply generating content, models are starting to act more like agents—that is, systems that can make decisions, use tools, and carry out multi-step tasks. o3 is one of the clearest public examples of this shift so far.

We’re also seeing this trend come to life in platforms like SmythOS, which lets developers build custom AI agents that can perform tasks across tools, APIs, and workflows.

SmythOS gives AI the structure and memory it needs to act across multiple steps, combining language understanding with action. It’s one of several emerging ecosystems turning powerful models into usable systems.

That’s the real breakthrough here. It’s about an AI that knows how to approach a challenge, plan its next move, and adjust along the way. That’s not superintelligence—but it might be the foundation that gets us there one day.

So while o3’s geoguessing might feel like a magic trick, what we’re really witnessing is the rise of AI as a thinking tool.

Conclusion: Awe, Caution, and What Comes Next

o3’s geoguessing skills show us just how far AI has come. In some moments, it matches or even beats human experts. But it’s not superintelligent. It’s specialized, structured, and still needs help to perform at its best.

What it really offers is a look at the next generation of AI tools—ones that reason, plan, and adapt. That shift matters more than any single guessing game. It’s a sign that AI is moving from a passive assistant to an active partner in problem-solving.

If you’re building with AI and want to tap into this new agentic future, SmythOS is where to start. Use it to create your own smart workflows, connect tools, and deploy AI that thinks in steps—not just sentences.

Try SmythOS and see what your AI can do when it’s not just answering, but truly reasoning.

Article last updated on: Last updated: May 30, 2025

What Is o3 and What Can It Do?

The Geoguessing Process

o3 vs. The World in GeoGuessr

The Sam Patterson Match

Where o3 Has the Edge

Where Humans Still Win

Is o3 Superhuman at GeoGuessr?

Is This Superintelligence? Let’s Define It

What Superintelligence Really Means

o3 Isn’t That — And Real Users Know It

Not a Glimpse of the Future — Yet

Why o3 Feels Like Superintelligence Anyway

1. It Explains Its Thinking

2. It Uses Tools Like a Detective

3. It Pays Attention to Tiny Clues

4. The “Shock” of Progress

The Real Signal — Not Superintelligence, but a Super Tool

Conclusion: Awe, Caution, and What Comes Next

Alexander De Ridder

Explore All AI Trends Articles

DeepMind Creates AlphaEvolve: An AI Agent That Invents

GPT-4.1 in ChatGPT: What Changed, Who Benefits, and Why It Matters

Audible’s AI Gambit: Can AI Voices Really Change the Story?

HealthBench: Can We Finally Trust AI Doctors?

Ludus AI in Unreal Engine: Helpful Sidekick or Over‑Hyped Hologram?

What Tech Jobs Will Be Safe From AI—At Least for the Next 5-10 Years?

Ready to Scale Your Business with SmythOS?

Ludus AI in Unreal Engine: Helpful Sidekick or Over‑Hyped Hologram?