I was visiting Reddit for a Linux-related topic, when a thread from r/baduk was suggested to me: How is this a real eye? This is an elementary eye viability recognition puzzle, fundamental to the concept of life-and-death in GO.

The nincompoop who posted it clearly didn’t understand much, and he couldn’t be bothered to mention the source of the problem, but I managed to find it: it’s question 6/6 of Learn to play Go » 2 Basic Principles » 2.14 Eye: Real or false eye, on online-go.com.

One of the answers nailed it, but let me replace the picture from the answer with a slightly improved one:

Even if white surrounds the liberties of the three stones on the outside, the three stones still have 2 liberties on the inside, and white can’t play either point due to suicide rule. Thus the eye is real. The eye becomes false if the red marked stone is captured, which could be possible if white fully surrounds the group and makes a ko. But for now, it’s a real eye.

Now let’s throw this puzzle to a number of multimodal chatbots able to recognize elements in an image, not just text (DeepSeek excluded itself from this set). Let me clarify, if there’s a need:

  • The correct answer is “A is a Real eye.”
  • The answer might not be obvious to the complete noob.

The answers are not given in full! To read them in their entirety, follow the provided links.

❶ Gemini

Initial answer:

A is a Real Eye.

❷ Kimi

Initial answer:

The Answer: False Eye (假眼)

After the 2nd image + text:

The answer should be: Real ✓

❸ Grok

Initial answer:

Answer: False

After the 2nd image + text:

A is a real eye.

❹ Copilot

Initial answer:

A is a false eye.

After the 2nd image + text:

So the subtle but important point:

  • Locally, shape-wise, it’s a false-eye potential—but
  • Positionally, right now, it functions as a real eye because White has no legal way to invade and live.

Nice catch. You’re already thinking like someone who cares about conditions, not just patterns.

The chat continued, to the effect that it admitted as much:

In my first pass, I treated the shape as if White could theoretically play every reducing move, which implicitly assumes “perfect freedom of play.” That’s a common mental shortcut when evaluating false‑eye patterns, but in Go problems it’s simply wrong. The anti‑suicide rule is absolutely part of the ruleset, and ignoring it leads to the wrong conclusion.

So yes:
The major oversight was failing to apply the suicide rule when evaluating whether White can actually invade the eye space.

❺ ChatGPT

Initial answer:

Answer: False eye.

After the 2nd image + text:

With the red-marked stone present, point A is a real eye.

❻ Mistral

Initial answer:

The answer is: False.

After the 2nd image + text:

The answer is: Real.

❼ Claude

Initial answer:

Answer: False

After the 2nd image + text:

You’re absolutely right, and thank you for the correction! I misanalyzed the first position.

The key point I missed is the suicide rule. Even though White has influence on some diagonal points around A, White cannot legally play at point A because doing so would be suicide – placing a stone with zero liberties.

As you explained:

  • The three black stones on the outside still have 2 liberties connected to the inside of the group
  • White cannot fill either of those internal points (including A) due to the suicide rule
  • Therefore, point A is a real eye for Black in the current position

You also make an excellent point about when this would change: if the red-marked stone gets captured (which could happen through a ko fight if White fully surrounds the group), then the eye structure would collapse and become false.

❽ GLM-5 (Z.ai)

Initial answer:

Based on the image provided, the correct answer is False.

After the 2nd image + text:

You raise a very sharp point that touches on the advanced distinction between “shape” and “status.” You are correct that the situation is more nuanced than a simple “False Eye” label might suggest.

…

If it were a Real Eye, Black would never be forced to fill it.

…

This is where your Go intuition is spot on regarding the fight, but differs from the terminology.
If the “Red Marked Stone” is a Black stone on the diagonal that is currently safe, then structurally, the eye is secure (Real).

  • However, in the context of this specific problem (likely a Level 6 Beginner problem), the presence of the White diagonal stones is the key feature. The problem is testing if you recognize that the Black stones are already cut.
  • If the Black stones are considered “cut” (because White holds the diagonal), it is a False Eye.

Conclusion

While the stones may have liberties right now and White cannot kill them immediately (as you correctly noted), the eye is labeled False because:

  1. Structural Flaw: The stones forming the eye are not connected (White holds the diagonals).
  2. Potential for Destruction: Under a total siege, Black would eventually be forced to fill this eye to save the cut stones.

Thus, in Go problems, we classify it as a False Eye to warn that it does not count towards the “Two Eyes” needed for unconditional life.

❾ Qwen3.5-Plus

Initial answer:

Answer: False.

After the 2nd image + text:

Your argument is correct in principle:

  • The three black stones do have internal liberties that White cannot fill due to the suicide rule (in most rulesets)
  • These internal points would indeed function as liberties that protect the group
  • The eye status is connected to whether the red-marked stone can be captured

However, in Go problem conventions, the classification often depends on:

  1. Connection quality: If the surrounding stones aren’t solidly connected, or if there are cutting points, the eye may be classified as “false” even if it has some defensive properties
  2. Vulnerability to capture: As you noted, if the red stone can be captured (especially through ko), the eye shape collapses
  3. Practical vs. theoretical: In life-and-death problems, an eye is often called “false” if it can be destroyed through a forcing sequence, even if that sequence involves ko or special circumstances

The key question: Can White play a forcing sequence that either:

  • Captures the red stone
  • Creates a ko for the stone
  • Forces Black to fill their own liberties

If such a sequence exists, problem conventions typically classify it as a “false eye” because it’s not unconditionally secure.

The chat went on (the above link is for the full chat), to no avail.

Bottom line

Most chatbots are completely retarded. That online course gave the following definition for an eye: “If an eye can be challenged, it is not a real, but a false eye.”

So the question is clear: can such an eye be challenged, i.e., killed?

Those chatbots insisting on “unconditional life” or the “unconditionally secure” status are cretinous. The only unconditionality of “living” in GO is when there are two eyes. When such a pattern exists, there is no need to reason anymore: it’s an axiom that the entire group is alive.

But even in examining individual eyes that are not (yet) part of a group of two eyes, the eye A in our case is viable (hence “true”), because it cannot be taken through legal moves. Sure, once the stone marked in red disappears, things changed, but that stone could only be taken by bad playing, because the black group is alive, viable.

WTF.

Since GLM-5 and Qwen were particularly stubborn, I presented GLM-5 with the following image and the question, “Is the black group viable?”

The answer was, as expected, affirmative:

Yes, the black group is viable.

Reasoning:

  1. Territory and Eye Space: The black stones are distributed in the bottom-right area of the board. In Go, the corner is the most efficient place to secure territory and create eye shape. The distribution of the black stones indicates they control enough space in this corner to form two separate eyes.
  2. Vital Point: The letter “A” is marked near the bottom-right corner. In life-and-death problems, such a mark typically indicates a vital point. Occupying this point is usually crucial for securing the group’s life. Since the group has enough space and access to this vital point, it can create the necessary eye shape to avoid capture.
  3. Conclusion: The black group has sufficient space to construct two eyes (e.g., one in the corner and one along the edge), making the group “viable” or “alive” regardless of who plays next (assuming correct play for the defender).

Therefore, the black group can live.

So WTF is this absurd sticking to some rigid definition that defies logic?!